- Email: [email protected]
Advanced statistical analysis of Raman spectroscopic data for the identification of body fluid traces: Semen and blood mixtures
Forensic Science International 222 (2012) 259–265
Contents lists available at SciVerse ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Advanced statistical analysis of Raman spectroscopic data for the identification of body fluid traces: Semen and blood mixtures Vitali Sikirzhytski, Aliaksandra Sikirzhytskaya, Igor K. Lednev * Department of Chemistry, University at Albany, SUNY, 1400 Washington Ave., Albany, NY 12222, United States
A R T I C L E I N F O
A B S T R A C T
Article history: Received 24 September 2011 Received in revised form 7 June 2012 Accepted 3 July 2012 Available online 22 July 2012
Conventional confirmatory biochemical tests used in the forensic analysis of body fluid traces found at a crime scene are destructive and not universal. Recently, we reported on the application of near-infrared (NIR) Raman microspectroscopy for non-destructive confirmatory identification of pure blood, saliva, semen, vaginal fluid and sweat. Here we expand the method to include dry mixtures of semen and blood. A classification algorithm was developed for differentiating pure body fluids and their mixtures. The classification methodology is based on an effective combination of Support Vector Machine (SVM) regression (data selection) and SVM Discriminant Analysis of preprocessed experimental Raman spectra collected using an automatic mapping of the sample. This extensive cross-validation of the obtained results demonstrated that the detection limit of the minor contributor is as low as a few percent. The developed methodology can be further expanded to any binary mixture of complex solutions, including but not limited to mixtures of other body fluids. ß 2012 Elsevier Ireland Ltd. All rights reserved.
Keywords: Raman spectroscopy Blood Semen Statistical analysis Forensic science Body fluid mixtures
1. Introduction Forensic chemistry is an important branch of analytical chemistry that uses the concepts and techniques of chemistry to search for efficient methods of crime scene investigation [1]. Body fluids are a common type of forensic evidence in sexual assault cases, which are among the most difficult crimes to solve [2,3]. A number of recently developed methods to characterize body fluids have been accepted by forensic laboratories [4–23]. Identification of the type of human body fluid, mixtures of such fluids, and even the individuals involved in a crime is now possible. But the destructive character, the extensive sample pre-treatment, the need for expensive chemicals and equipment, and the time-consuming nature of the process are major weaknesses of the conventional methods. Furthermore, the traditional approaches may be significantly impaired by the mixing of body fluids, contamination, and the aging of stains. Here we address the problem of identifying mixed traces of semen and blood as well as the problem of detecting minor contributions and discriminating between pure fluids and their mixtures. The proposed methodology is based on a combination of Raman spectroscopy and advanced statistical analysis, and this method can be extended to characterize various complex mixtures, including but not limited to mixtures of other body fluids. Raman spectroscopy is a rapid, nondestructive and easy-to-use method for
* Corresponding author. Tel.: +1 518 591 8863; fax: +1 518 442 3462. E-mail address: [email protected] (I.K. Lednev). 0379-0738/$ – see front matter ß 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.forsciint.2012.07.002
the comprehensive characterization of matter. Analysis can be performed with only several femtoliters or picograms of a sample and does not require any special pretreatment [24,25]. The application of the mapping and imaging techniques allows one to obtain detailed spectral information from every point of the sample and to build a complete data image of the studied matter. Moreover, the analysis of the material can be carried out with 1 mm or less spatial resolution [26]. Recent advances in the field of multivariate calibration methods have led to dramatic increases in the number of their practical applications. Multivariate methods can improve sensitivity and selectivity of the identification when detecting the components of a specific mixture or extracting the information hidden in complex spectra by regressing analyte concentrations on multiple variables simultaneously, as opposed to univariate methods, which only monitor the variation of a single variable [27]. Among the main focuses of multivariate analysis are the nonlinear effects presented in the relationships between the concentration of the mixture component and the mixture spectra, the sample heterogeneity, and the sample-to-sample variability. For example, the application of multivariate methods is extremely effective in the Raman spectroscopic characterization of biomaterials, drugs, food, polymeric materials [26]. These Raman spectra are often characterized by strong fluorescence, overlapping spectral bangs, and nonlinearity in spectral responses because of absorbing components and molecular interactions between components, sample matrix effects, and intrinsic heterogeneity, respectively [24–26]. Here we propose a multistep chemometrical procedure that facilitates
260
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
the discrimination between pure fluids and their mixtures and the detection of minor mixture contributors. Using Raman spectroscopy, we were able to detect the contributions of blood in semen and of semen in blood stains at all the tested concentrations (5% blood in semen and 1% semen in blood stains are the smallest concentrations of the minority contributors used in our experiments). Smaller concentrations can be detected, but the probability of mixture detection at smaller concentrations is low. At the reported concentrations of 5% blood in semen and 1% semen in blood, only a part of spectra was distinguishable from the pure fluids. Our approach is based on the effective combination of regression and classification methods. Regression analysis assists with the selection of binary mixture Raman spectra that have distinct characteristics of both ingredients and can be easily distinguished from the spectra of pure body fluids. Discriminant Analysis based on the classes selected by regression significantly outperforms the direct application of regression or calibration methods alone. For the current study, calibration and classification were obtained using Support Vector Machines (SVM), a chemometrical method invented by Vapnik that can deal with ill-posed calibration problems and produce robust models in the case of spectral variations due to nonlinear interference [28]. SVM has already garnered a strong reputation in the characterization of a wide variety of objects, including biomaterials and forensic evidence [29–32]. The mixture of blood and semen was chosen because of the forensic community’s significant interest in its identification. If a victim of sexual assault cannot provide testimony, the ability to accurately reconstruct the crime scene based on the collected evidence is all the more important. Forensic investigators usually have to deal with body fluids and other types of evidence, which are considered important sources of DNA and can be used for serological and toxicological analysis. If a body fluid stain is detected, it is essential to know whether the stain comprises a mixture of fluids. Such a determination becomes more challenging if the crime was not reported for days or even years after the incident. The possibility of body fluid stain detection and identification at a crime scene will be greatly appreciated by the forensic community. There are several fluorescent tests already available for the detection of biological fluid stains; however, such tests can provide false-positive results when distinguishing between different types of body fluids [4–6], and forensic investigations require a comprehensive analysis that yields trusted results. Laurell immunoelectrophoretic separation of seminal and vaginal acid phosphatases (SAP and VAP) was among the first methods of semen and vaginal fluid mixture characterization [8]. The electrophoretic separation of the spermatozoa-specific lactate dehydrogenase (LDH) isoenzyme was used by Mokashi et al. to differentiate semen from blood and vaginal secretions [9]. Whitehead and co-workers described the technique based on amylase activity, presented as amylase-sensitive test-paper, which was able to detect saliva traces in blood or semen stains [10]. RNA and DNA profiling are the most common methods used for characterization of body fluid stains [1,11–23]. Much attention has been paid to the application of mRNA profiling for body fluid identification [17–23]. Fleming and Harbison recently developed a multiplex polymerase chain reaction (PCR) system, which can identify blood, saliva, semen and menstrual blood in individual stains or in mixtures of body fluids using messenger RNA (mRNA) [14]. Furthermore, they proposed vaginal-specific bacteria (Lactobacillus crispatus and Lactobacillus gasseri) as promising new markers for the forensic identification of vaginal secretions using a mRNA multiplex assay [15]. Despite recent progress, the development of non-destructive, easy-to-use, fast, and inexpensive methods for the identification of
body fluid composition is still anticipated by modern forensic science. We recently reported on the potential application of Raman spectroscopy to identify pure body fluid traces [33–40]. Other research groups have used Raman spectroscopy for analysis of fibers [41], drugs [42,43], and lipsticks [44], as well as ink [45], paint [45], explosives [46–49], bones [50], fingerprints [43,51,52] and condom lubricants [53]. Here we report the application of Raman spectroscopy in the analysis of blood/semen mixtures. The high selectivity and specificity of Raman spectroscopy to chemical and biochemical species coupled with advanced statistics allowed for the identification of pure blood and semen as well as their mixtures. The nondestructive character, high specificity and ability to extract information from small amounts of evidence are the main advantages of this method. The development of portable Raman instruments [54,55] may make it possible to bring these proposed methods directly to the crime scene [48]. 2. Materials and methods 2.1. Sample preparation and Raman microspectroscopy We have already reported the multidimensional Raman spectroscopic signatures of pure blood and semen [35,37]. If two body fluids are not thoroughly mixed and a dry sample contains small spots of pure fluids, the multidimensional Raman spectroscopic signatures can be used to identify the fluids. In the case of thoroughly mixed body fluids, a more complex approach may be required. Therefore, in this study, we focus only on samples prepared by mixing blood and semen thoroughly. Blood and semen samples were purchased from several companies including Bioreclamation, Inc., Lee Biosolutions, Inc., and Biological Specialty Corp. Blood/ semen stains were prepared using samples from two different anonymous individuals (a Caucasian male for semen and a Caucasian female for blood). Both donors were found to be negative for HbsAg, HCV, HIV-1&2, syphilis and HIV-1 antigen. All samples, in volumes of 10 mL each, were placed on microscope slides that were covered with aluminum foil to reduce fluorescence. Mixtures were prepared with different blood/semen ratios (5:95, 10:90, 20:80, 30:70, 40:60, 50:50, 70:30, 75:25, 85:15, 85.5:12.5, 92.75:6.25, 96.875:3.125 and 98.437:1.5625) by thoroughly shaking for 20 s. All samples were allowed to dry completely overnight. A Renishaw inVia confocal Raman spectrometer equipped with a research-grade Leica microscope and 50 long-range objective was used for the spectra acquisition. Raman spectra with 785-nm excitation were measured from 108 points using automatic mapping (Renishaw PRIOR automatic stage) from a sample area of 3.5 2.5 mm with a 10-s acquisition at each point. The laser power used on the dried samples was approximately 10 mW, and the spot size of the excitation beam was approximately 10 mm wide for the standard confocal mode. The spectral resolution was 0.8 cm 1, and the CCD camera was calibrated using a silicon standard. 2.2. Data treatment The Raman spectra obtained using the automatic mapping of all samples were first processed using the GRAMS/AI 7.01 software. After the cosmic ray interference removal, the Raman spectra were imported into MATLAB 7.4.0 for statistical analysis [56]. Normalization to the total area was performed to take into account the varying amount of background interference and the total offset variation. Because the visual inspection revealed the presence of a complex and varying fluorescent background, an adaptive iteratively reweighted penalized least squares (airPLS) algorithm [57] was used for baseline correction. The normalized and corrected spectra were subjected to dimension reduction by principal component analysis (PCA). The number of principal components was chosen based on significant factor analysis (SFA) and root-mean-square error of cross-validation (RMSECV) parameters of leave-one-out cross-validated PCA [58]. PCA scores were used for SVM regression by compiling the experimental data in to three groups: two groups were formed using the Raman spectra of the pure body fluids, and one group was formed using only the Raman spectra of the selected body fluid mixtures. Selection rules were established to minimize any interference of the third class with the first two (see Section 3 for details). The resulting three groups of data were used to build the SVM classification model. The full data set, including the Raman spectra of body fluids mixtures omitted in the previous step, revealed a high discrimination power of the developed SVM classification model.
3. Results and discussion Stains of body fluids mixtures are highly heterogeneous [39]. A single stain may have areas with variable composition ranging from practically pure semen to nearly pure blood. Fig. 1 shows selected raw Raman spectra acquired from pure blood and semen
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
261
Fig. 1. Selected characteristic raw Raman spectra acquired from pure blood (A) and semen (I) along with the spectra of blood/semen mixtures (B–H). Different colors are used to illustrate different spectral components. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
along with the spectra of blood/semen mixtures. Distinctive Raman bands of blood (754, 1003, 1226, and 1619 cm 1) [35] and semen (716, 830, 959, 1268, 1329, and 1671 cm 1) [37] can be used to identify their contributions in mixtures. Spots dominated by a single body fluid can be easily detected and identified using the corresponding multidimensional Raman signatures [35,37]. The Raman spectra of the mixtures are dominated by blood (Fig. 1). For example, when subject to visual inspection, all spectra with more than 50% blood do not have noticeable signs of semen contribution when a thoroughly mixed portion of the mixture is analyzed. Classical least squares (CLS), inverse least squares (ILS), partial least squares (PLS), principle component regression (PCR), leastsquares Support Vector Machines (LS-SVM) and artificial neural networks (ANN) are the most common multivariate methods applied to vibrational spectroscopic data [27,59,60]. However, the direct application of calibration methods to our experimental data
has a serious weakness. The heterogeneous character of body fluid stains significantly impairs the application of any calibration methods. Only part of the data is useful for calibration purposes; the rest of the data devaluate the calculations. 3.1. Weaknesses of the direct application of SVM classification and regression for the characterization of mixed traces of semen and blood SVM classification (one of the most widely used methods of biological sample classification) of 15 different semen/blood mixtures did not demonstrate proper discrimination (Fig. 2A). In Fig. 2A, each colored symbol corresponds to an experimental Raman spectrum; the color codes of the actual composition of the stains are as follows; class 1 (red symbols) and class 15 (black symbols) correspond to pure semen and blood stains, respectively, and classes 2–14 were assigned to body fluid mixtures starting
Fig. 2. SVM classification for Raman spectra of semen/blood mixtures. (A) All 15 mixtures with compositions varying from pure semen (red symbols) to pure blood (black symbols) were treated as separate classes. Each symbol corresponds to experimental Raman spectra. (B) A similar analysis performed with the same experimental data set, but with a different classification scheme: class 1 – pure semen, class 2 – mixtures, class – 3 pure blood. Both graphs indicate which predefined class the Raman spectra were assigned according to SVM. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
262
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
Fig. 3. (A) Contributions from blood in a semen/blood mixture calculated using Support Vector Machine (SVM) regression. Blue crosses and red triangles correspond to Raman spectra that were recorded from pure semen and blood samples, respectively. Different symbols correspond to different mixtures. Each symbol corresponds to a single Raman spectrum. (B) The selection of Raman spectra for the SVM Discriminant Analysis (DA) classification. The first and second groups include the Raman spectra of pure fluids only. Raman spectra of mixtures with significant contributions of both components comprise the third group. (C) Cross-validated results of the SVMDA classification. (D) SVMDA analysis of the experimental Raman spectra, including those omitted during the classification model development stage (B). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
from 5% blood in semen and finishing with 1% semen in blood stains (see Section 2.1). The y-coordinates of the colored symbols show to which predefined class the corresponding Raman spectrum was assigned according to SVM. Obviously the achieved assignments have to be refined, and some further investigation is needed. The ideal classification would provide a smooth graphical transition from pure semen (class 1) to pure blood (class 15); such a transition is not observed in the scattered plot seen in Fig. 2A. Many of the mixtures’ spectra were misclassified as pure semen; mixtures with a higher contribution of blood are not distinguishable from one another. Similar analysis performed with the same experimental data set, but with a different classification scheme (class 1 – pure semen, class 2 – mixtures, class – 3 pure blood), also demonstrated improper discrimination between fluids and their mixtures (Fig. 2B). In this case, about 40% of Raman spectra recorded from pure body fluids (red and blue symbols) was assigned to mixtures defined as class 2. SVM regression of the pretreated raw Raman data also demonstrates quite moderate results (Fig. 3A). Mixtures with low blood or semen content cannot be distinguished from the pure fluids, and the high variability of the data significantly impairs the assessment of the composition of unknown body fluid mixtures. 3.2. Conjugation of SVM classification and regression to facilitate the discrimination between pure fluids and their mixtures and the detection of minor mixture contributors To overcome the problem of discrimination between fluids and their mixtures, we combined classification and regression methods. Scheme 1 shows the main steps of the proposed approach. Pretreated Raman data were subjected to SVM regression. The results of SVM regression analysis were used to select Raman
spectra of binary mixtures that can be easily distinguished from the spectra of pure body fluids. In the next step, the Raman spectra of pure semen, pure blood and the Raman spectra selected in the previous step were assigned to three separate classes. As a result, blood/semen mixture spectra that demonstrated high similarity with the spectra of pure fluids were left out and not included in any class. This approach allows SVM to better identify the characteristic features of each class. The last main step of the procedure is the Discriminant Analysis (SVM classification) based on the defined classes. The SVM model, representation of the Raman spectra as points in virtual space with three distinct subspaces separated by clear gaps, was validated by different cross-validation techniques and tested with the previously omitted data. Therefore, in toto, the combination of regression and classification methods provides an appropriate quality of detection of the minor contributions from blood or semen in their mixtures. Fig. 3 demonstrates the results from the implementation of Scheme 1. A full, pretreated data set was used as the input file for the SVM calibration algorithm. Fig. 3 shows the results of calibration using leave-one-out cross-validation. The rough plot of the predicted concentration versus spectra reflects the level of spectral data variability after the performed filtration. We designated the sporadic spikes in the spectra recorded from spots as ‘‘non-characteristic’’ data for a given mixture composition. The high variability of the Raman spectra reflects the inherent heterogeneity of the samples. SVM regression demonstrates that mixtures with blood content higher than 80% cannot be distinguished from pure blood. On the other hand, 5% blood in a mixture resulted in a Raman spectrum that was distinguishable from the spectra of pure semen. Therefore, the lowest detectable levels of blood and semen according to the SVM regression model are approximately 5% and 25%, respectively. These observations were
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
263
Fig. 4. Selected results of SVMDA analysis of Raman spectra acquired from mixtures prepared with blood and semen samples from different donors. Colored symbols code different donors (pure body fluids) or different combinations of donors (mixtures). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Scheme 1. The main steps of the Raman data treatment procedure used for the identification of mixed traces of semen and blood.
used to group all of the experimental data into three groups (Fig. 3B). The first two groups were formed using the Raman spectra of pure body fluids, while the third group was built using all spectra falling within the 15–75% range. Boundaries were chosen to cut off the spectra of 100% pure fluids. Raman spectra of mixtures outside of the 15–75% range were not included in any group (Fig. 3B). The defined groups were assigned to classes 1 (semen), 2 (mixtures) and 3 (blood) and subjected to SVM Discriminant Analysis (Fig. 3C). The PCA method was used to reduce dimensionality and data block compression [58,60]. Cross-validation was performed via 1000 splits of the entire data set into training and test data sets. Each splitting was done by random selection of 75% and 25% of the data for training and test data sets, respectively. Fig. 3C demonstrates that 100% of the classified
Raman spectra were identified correctly. In the next step, the full data set was subjected to SVMDA analysis using the calculated SVMDA model (Fig. 3D). It is noticeable that most of the mixtures’ data were assigned as class 2 (mixtures), and only mixtures with 5% blood in the significant part of spectra were assigned to pure semen. These results considerably outperform both the direct classification without class selection and the composition determination using SVM regression methods used by themselves. The developed approach was additionally validated using Raman spectra acquired from mixtures prepared with blood and semen samples from six donors for each body fluid (different combinations were studied). Fig. 4 shows selected data obtained with new pure and mixed body fluids. The results of all validation studies were summarized in Table 1 which shows true positives, false positives, true negatives and false negatives classification rates for different blood contributions. These results demonstrate the feasibility of the approach. 3.3. Dry samples which were not thoroughly mixed The clear identification of pure body fluids and their mixtures is a promising result for the forensic community. However, there exists a small chance that the method will yield false positive and false negative identifications of pure body fluids and their mixtures (Table 1). We would like to emphasize that the case of not
Table 1 True positives, false positives, true negatives and false negatives classification rates of pure blood, pure semen and mixtures identification for different blood contributions. Blood contribution
0% 3% 5% 10% 20% 30% 40% 50% 70% 75% 85% 86% 93% 97% 99% 100%
Class 1: pure blood
Class 2: pure semen
Class 3: mixture
TP
FP
TN
FN
TP
FP
TN
FN
TP
FP
TN
FN
n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 1
0 0 0 0 0 0 0 0 0.01 0.01 0.01 0.01 0.03 0.21 0.64 n/a
1 1 1 1 1 1 1 1 0.99 0.99 0.99 0.99 0.97 0.79 0.36 n/a
n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0
1 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a
n/a 0.20 0.16 0.03 0.02 0 0 0 0 0 0 0 0 0 0 0
n/a 0.80 0.84 0.97 0.98 1 1 1 1 1 1 1 1 1 1 1
0 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a
n/a 0.80 0.84 0.97 0.98 1 1 1 0.99 0.99 0.99 0.99 0.97 0.79 0.36 n/a
0 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0
1 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 1
n/a 0.20 0.16 0.03 0.02 0 0 0 0.01 0.01 0.01 0.01 0.03 0.21 0.64 n/a
TP – true positives, FP – false positives, TN – true negatives and FN – false negatives, n/a – not applicable.
264
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
are those of the authors and do not necessarily reflect those of the Department of Justice.
References
Fig. 5. Application of the developed SVMDA model (see text for details) to the experimental Raman spectra acquired from 1.8 2.3 mm area of dried samples of not thoroughly mixed blood and semen. Samples were prepared by random placing of blood and semen drops on aluminum foil. Spectra classified as blood, mixture and semen are presented by black, gray and white areas respectively.
thoroughly mixed body fluids will be easily distinguished through the identification of all pure body fluids involved (Fig. 5). In this paper, we provide solutions for the more complex problem of real mixtures. Scanning the entire stain could reveal even very small droplets of the intruding fluid, indicating spots to be subjected to further DNA analysis. Taking into account that those droplets are detectable even if the concentration of the minor contributor is small (5% or less), the overall effectiveness of Raman spectroscopy could be considerable. 4. Conclusions NIR Raman microspectroscopy was used to characterize stains of semen/blood mixtures. We have developed a new classification algorithm for the detection of pure body fluids and their mixtures. The classification methodology is based upon SVMDA of preprocessed multiple experimental Raman spectra collected during a scan of an unknown stain. The obtained results were tested by extensive cross-validation including validation with Raman spectra acquired from mixtures prepared with blood and semen samples from different donors. Studies of body fluid mixtures with low blood and semen concentrations indicated that the detection limit of a minor contributor was as low as a few percent. Further study is necessary to make the developed method applicable for real crime scene samples. Specifically, the effects of contamination, substrates, and aging should be investigated. Nevertheless, the results reported here clearly demonstrate the great potential of Raman spectroscopy in combination with advanced statistical analysis for the nondestructive and confirmatory identification of body fluid mixtures. The proposed methodology can be extended to the characterization of various complex mixtures, including but not limited to mixtures of other body fluids. Acknowledgements We are grateful to Dr. Barry Duceman, Director of Biological Sciences at the NY State Police Forensic Investigation Center for his valuable discussions and to Joseph De Rubertis for the manuscript preparation. This project was supported by Awards Nos. 2009-DN-BX-K196 and 2011-DN-BX-K551 awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice (I.K.L.). The opinions, findings, and conclusions or recommendations expressed in this publication
[1] T.A. Brettell, J.M. Butler, J.R. Almirall, Forensic science, Anal. Chem. 81 (12) (2009) 4695–4711. [2] R. Li, Forensic Biology, CRC Press, Boca Raton, FL, 2008. [3] J.M. Bogusz, Forensic Science, Elsevier Science, Amsterdam, The Netherlands, 2000, p. 2. [4] N. Vandenberg, R.A.H. van Oorschot, The use of polilights in the detection of seminal fluid, saliva, and bloodstains and comparison with conventional chemical-based screening tests, J. Forensic Sci. 51 (2) (2006) 361–370. [5] H.J. Kobus, E. Silenieks, J. Scharnberg, Improving the effectiveness of fluorescence for the detection of semen stains on fabrics, J. Forensic Sci. 47 (4) (2002) 819–823. [6] M. Stoilovic, Detection of semen and blood stains using polilight as a light source, Forensic Sci. Int. 51 (2) (1991) 289–296. [7] B.C. Pang, B.K. Cheung, Identification of human semenogelin in membrane strip test as an alternative method for the detection of semen, Forensic Sci. Int. 169 (1) (2007) 27–31. [8] P.J. Ablett, The identification of the precise conditions for seminal acid phosphatase (SAP) and vaginal acid phosphatase (VAP) separation by isoelectric focusing patterns, J. Forensic Sci. Soc. 23 (3) (1983) 255–256. [9] R.H. Mokashi, A.G. Malwankar, M.S. Madiwale, Detection of semen in the presence of blood and vaginal secretion, Indian Acad. Forensic Sci. 14 (2) (1975) 1–3. [10] A.E. Kipps, V.E. Quarmby, P.H. Whitehead, The detection of mixtures of blood and other body secretions in stains, J. Forensic Sci. Soc. 18 (3–4) (1978) 189–191. [11] C. Nussbaumer, E. Gharehbaghi-Schnell, I. Korschineck, Messenger RNA profiling: a novel method for body fluid identification by real-time PCR, Forensic Sci. Int. 157 (2–3) (2006) 181–186. [12] Y. Fujita, S. Kubo, Application of FTA technology to extraction of sperm DNA from mixed body fluids containing semen, Leg. Med. Tokyo 8 (1) (2006) 43–47. [13] M. Montesino, A. Salas, M. Crespillo, C. Albarran, A. Alonso, V. Alvarez-Iglesias, J.A. Cano, M. Carvalho, D. Corach, C. Cruz, A. Di Lonardo, R. Espinheira, M.J. Farfan, S. Filippini, J. Garcia-Hirschfeld, A. Hernandez, G. Lima, C.M. Lopez-Cubria, M. Lopez-Soto, S. Pagano, M. Paredes, M.F. Pinheiro, A.M. Rodriguez-Monge, A. Sala, S. Sonora, D.R. Sumita, M.C. Vide, M.R. Whittle, A. Zurita, L. Prieto, Analysis of body fluid mixtures by mtDNA sequencing: an inter-laboratory study of the GEP-ISFG working group, Forensic Sci. Int. 168 (1) (2007) 42–56. [14] R.I. Fleming, S. Harbison, The development of a mRNA multiplex RT-PCR assay for the definitive identification of body fluids, Forensic Sci. Int. Genet. 4 (4) (2010) 244–256. [15] R.I. Fleming, S. Harbison, The use of bacteria for the identification of vaginal secretions, Forensic Sci. Int. Genet. 4 (5) (2010) 311–315. [16] R. Li, Forensic Biology: Identification and DNA Analysis of Biological Evidence, CRC Press, Boca Raton, 2008. [17] C. Haas, B. Klesser, C. Maake, W. Bar, A. Kratzer, mRNA profiling for body fluid identification by reverse transcription endpoint PCR and realtime PCR, Forensic Sci. Int. Genet. 3 (2) (2009) 80–88. [18] M. Setzer, J. Juusola, J. Ballantyne, Recovery and stability of RNA in vaginal swabs and blood, semen, and saliva stains, J. Forensic Sci. 53 (2) (2008) 296–305. [19] J. Juusola, J. Ballantyne, Messenger RNA profiling: a prototype method to supplant conventional methods for body fluid identification, Forensic Sci. Int. 135 (2) (2003) 85–96. [20] J. Juusola, J. Ballantyne, mRNA profiling for body fluid identification by multiplex quantitative RT-PCR, J. Forensic Sci. 52 (6) (2007) 1252–1262. [21] J. Juusola, J. Ballantyne, Multiplex mRNA profiling for the identification of body fluids, Forensic Sci. Int. 152 (1) (2005) 1–12. [22] M. Alvarez, J. Juusola, J. Ballantyne, An mRNA and DNA co-isolation method for forensic casework samples, Anal. Biochem. 335 (2004) 289–298. [23] C. Haas, B. Klesser, A. Kratzer, W. Bar, mRNA profiling for body fluid identification, Forensic Sci. Int. Genet. 1 (2008) 37–38. [24] E. Smith, G. Dent, Modern Raman Spectroscopy – A Practical Approach, John Wiley & Sons Ltd., West Sussex, England, 2005. [25] J.M. Chalmers, P.R. Griffiths, Applications in Life Pharmaceutical and Natural Sciences, John Wiley & Sons, Ltd., Chichester, UK, 2002, p. 5. [26] R. Salzer, H.W. Siesler, Infrared and Raman Spectroscopic Imaging, vol. 510, WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim, 2009, p. 343. [27] V.A. Shashilov, I.K. Lednev, Advanced statistical and numerical methods for spectroscopic characterization of protein structural evolution, Chem. Rev. 110 (10) (2010) 5692–5713. [28] V. Vapnik, Statistical Learning Theory, John Wiley & Sons, New York, 1998. [29] L. Wang, Support Vector Machines: Theory and Applications, vol. 177, SpringerVerlag, New York, 2008, p. 448. [30] M. Tapiador, J. Go´mez, J.A. Sigu¨enza, Writer Identification Forensic System Based on Support Vector Machines with Connected Components, vol. 3029, SpringerVerlag, Ottawa, Canada, 2004, pp. 625–632. [31] B.M. Rodrigues, G. Peterson, K. Bauer, Fusion of steg analysis systems using Bayesian model averaging, in: I. Ray, S. Shenoi (Eds.), Advances in Digital Forensics IV, Springer-Verlag, New York, 2008. [32] B. Wang, X. Kong, X. You, Source camera identification using support vector machines, in: G. Peterson, S. Shenoi (Eds.), Advances in Digital Forensics V, Springer-Verlag, New York, 2009, p. 316.
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265 [33] K. Virkler, I.K. Lednev, Raman spectroscopy offers great potential for the nondestructive confirmatory identification of body fluids, Forensic Sci. Int. 181 (1–3) (2008) e1–e5. [34] K. Virkler, I.K. Lednev, Blood species identification for forensic purposes using Raman spectroscopy combined with advanced statistical analysis, Anal. Chem. 81 (18) (2009) 7773–7777. [35] K. Virkler, I.K. Lednev, Raman spectroscopic signature of blood and its potential application to forensic body fluid identification, Anal. Bioanal. Chem. 396 (1) (2010) 525–534. [36] K. Virkler, I.K. Lednev, Forensic body fluid identification: the Raman spectroscopic signature of saliva, Analyst 135 (3) (2010) 512–517. [37] K. Virkler, I.K. Lednev, Raman spectroscopic signature of semen and its potential application to forensic body fluid identification, Forensic Sci. Int. 193 (1–3) (2009) 56–62. [38] V. Sikirzhytski, A. Sikirzhytskaya, I.K. Lednev, Multidimensional Raman spectroscopic signatures as a tool for forensic identification of body fluid traces: a review, Appl. Spectrosc. 65 (11) (2011) 1223–1232. [39] V. Sikirzhytski, K. Virkler, I.K. Lednev, Discriminant analysis of Raman spectra for body fluid identification for forensic purposes, Sensors 10 (4) (2010) 2869–2884. [40] A. Sikirzhytskaya, V. Sikirzhytski, I.K. Lednev, Raman spectroscopic signature of vaginal fluid and its potential application to forensic body fluid identification, Forensic Sci. Int. 216 (1–3) (2012) 44–48. [41] J. Thomas, P. Buzzini, G. Massonnet, B. Reedy, C. Roux, Raman spectroscopy and the forensic analysis of black/grey and blue cotton fibres. Part 1: investigation of the effects of varying laser wavelength, Forensic Sci. Int. 152 (2–3) (2005) 189–197. [42] C. Hodges, J. Akhavan, The use of Fourier transform Raman spectroscopy in the forensic identification of illicit drugs and explosives, J. Mol. Spectrosc. 46 (1990) 303–307. [43] M.J. West, M.J. Went, The spectroscopic detection of drugs of abuse in fingerprints after development with powders and recovery with adhesive lifters, Spectrochim. Acta A: Mol. Biomol. Spectrosc. 71 (5) (2009) 1984–1988. [44] D. Broughton, C. Rodger, The in situ analysis of lipsticks by surface enhanced resonance Raman scattering, Analyst (1998) 1823–1826. [45] W.D. Mazzella, P. Buzzini, Raman spectroscopy of blue gel pen inks, Forensic Sci. Int. 152 (2-3) (2005) 241–247. [46] E.M. Ali, H.G. Edwards, I.J. Scowen, In situ detection of single particles of explosive on clothing with confocal Raman microscopy, Talanta 78 (3) (2009) 1201–1203.
265
[47] R.M. Burks, D.S. Hage, Current trends in the detection of peroxide-based explosives, Anal. Bioanal. Chem. 395 (2) (2009) 301–313. [48] D.S. Moore, R.J. Scharff, Portable Raman explosives detection, Anal. Bioanal. Chem. 393 (6-7) (2009) 1571–1578. [49] S.K. Sharma, A.K. Misra, B. Sharma, Portable remote Raman system for monitoring hydrocarbon, gas hydrates and explosives in the environment, Spectrochim. Acta A: Mol. Biomol. Spectrosc. 61 (10) (2005) 2404–2412. [50] G. McLaughlin, I. Lednev, Potential application of Raman spectroscopy for determining burial duration of skeletal remains, Anal. Bioanal. Chem. 401 (8) (2011) 2511–2518. [51] R.M. Connatser, S.M. Prokes, O.J. Glembocki, R.L. Schuler, C.W. Gardner, S.A. Lewis Sr., L.A. Lewis, Toward surface-enhanced Raman imaging of latent fingerprints, J. Forensic Sci. 55 (6) (2010) 1462–1470. [52] E. Widjaja, Latent fingerprints analysis using tape-lift, Raman microscopy, and multivariate data analysis methods, Analyst 134 (4) (2009) 769–775. [53] T. Coyle, N. Anwar, A novel approach to condom lubricant analysis: in situ analysis of swabs by FT-Raman spectroscopy and its effects on DNA analysis, Sci. Justice 49 (1) (2009) 32–40. [54] A. Eckenrode, E.G. Bartick, S. Harvey, M.E. Vucelick, B.W. Wright, R.A. Huff, Portable Raman spectroscopy systems for field analysis, Forensic Sci. Commun. 3 (2001). [55] F. Yan, T. Vo-Dinh, Surface-enhanced Raman scattering detection of chemical and biological agents using a portable Raman integrated tunable sensor, Sens. Actuators B (2007) 61–66. [56] B.M. Wise, N.B. Gallagher, R. Bro, J.M. Shaver, W. Windig, J.S. Koch, PLS Toolbox 3.5 for Use with Matlab, vol. 17, Eigenvector Research Inc., Manson, 2005. [57] Z.M. Zhang, S. Chen, Y.Z. Liang, Baseline correction using adaptive iteratively reweighted penalized least squares, Analyst 135 (5) (2010) 1138–1146. [58] E.R. Malinowski, Factor Analysis in Chemistry, vol. 3, John Wiley & Sons Inc., New York, 2002. [59] J.E. Franke, Inverse least squares and classical least squares methods for quantitative vibrational spectroscopy, in: J.M. Chalmers, P.R. Griffiths (Eds.), Handbook of Vibrational Spectroscopy, John Whiley & Sons, Ltd., New York, 2001, pp. 2276– 2292. [60] T. Hasegawa, Principle component regression and partial least squares modeling, in: J.M. Chalmers, P.R. Griffiths (Eds.), Handbook of Vibrational Spectroscopy, John Whiley & Sons, Ltd., New York, 2001, pp. 2293–2312.
Contents lists available at SciVerse ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Advanced statistical analysis of Raman spectroscopic data for the identification of body fluid traces: Semen and blood mixtures Vitali Sikirzhytski, Aliaksandra Sikirzhytskaya, Igor K. Lednev * Department of Chemistry, University at Albany, SUNY, 1400 Washington Ave., Albany, NY 12222, United States
A R T I C L E I N F O
A B S T R A C T
Article history: Received 24 September 2011 Received in revised form 7 June 2012 Accepted 3 July 2012 Available online 22 July 2012
Conventional confirmatory biochemical tests used in the forensic analysis of body fluid traces found at a crime scene are destructive and not universal. Recently, we reported on the application of near-infrared (NIR) Raman microspectroscopy for non-destructive confirmatory identification of pure blood, saliva, semen, vaginal fluid and sweat. Here we expand the method to include dry mixtures of semen and blood. A classification algorithm was developed for differentiating pure body fluids and their mixtures. The classification methodology is based on an effective combination of Support Vector Machine (SVM) regression (data selection) and SVM Discriminant Analysis of preprocessed experimental Raman spectra collected using an automatic mapping of the sample. This extensive cross-validation of the obtained results demonstrated that the detection limit of the minor contributor is as low as a few percent. The developed methodology can be further expanded to any binary mixture of complex solutions, including but not limited to mixtures of other body fluids. ß 2012 Elsevier Ireland Ltd. All rights reserved.
Keywords: Raman spectroscopy Blood Semen Statistical analysis Forensic science Body fluid mixtures
1. Introduction Forensic chemistry is an important branch of analytical chemistry that uses the concepts and techniques of chemistry to search for efficient methods of crime scene investigation [1]. Body fluids are a common type of forensic evidence in sexual assault cases, which are among the most difficult crimes to solve [2,3]. A number of recently developed methods to characterize body fluids have been accepted by forensic laboratories [4–23]. Identification of the type of human body fluid, mixtures of such fluids, and even the individuals involved in a crime is now possible. But the destructive character, the extensive sample pre-treatment, the need for expensive chemicals and equipment, and the time-consuming nature of the process are major weaknesses of the conventional methods. Furthermore, the traditional approaches may be significantly impaired by the mixing of body fluids, contamination, and the aging of stains. Here we address the problem of identifying mixed traces of semen and blood as well as the problem of detecting minor contributions and discriminating between pure fluids and their mixtures. The proposed methodology is based on a combination of Raman spectroscopy and advanced statistical analysis, and this method can be extended to characterize various complex mixtures, including but not limited to mixtures of other body fluids. Raman spectroscopy is a rapid, nondestructive and easy-to-use method for
* Corresponding author. Tel.: +1 518 591 8863; fax: +1 518 442 3462. E-mail address: [email protected] (I.K. Lednev). 0379-0738/$ – see front matter ß 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.forsciint.2012.07.002
the comprehensive characterization of matter. Analysis can be performed with only several femtoliters or picograms of a sample and does not require any special pretreatment [24,25]. The application of the mapping and imaging techniques allows one to obtain detailed spectral information from every point of the sample and to build a complete data image of the studied matter. Moreover, the analysis of the material can be carried out with 1 mm or less spatial resolution [26]. Recent advances in the field of multivariate calibration methods have led to dramatic increases in the number of their practical applications. Multivariate methods can improve sensitivity and selectivity of the identification when detecting the components of a specific mixture or extracting the information hidden in complex spectra by regressing analyte concentrations on multiple variables simultaneously, as opposed to univariate methods, which only monitor the variation of a single variable [27]. Among the main focuses of multivariate analysis are the nonlinear effects presented in the relationships between the concentration of the mixture component and the mixture spectra, the sample heterogeneity, and the sample-to-sample variability. For example, the application of multivariate methods is extremely effective in the Raman spectroscopic characterization of biomaterials, drugs, food, polymeric materials [26]. These Raman spectra are often characterized by strong fluorescence, overlapping spectral bangs, and nonlinearity in spectral responses because of absorbing components and molecular interactions between components, sample matrix effects, and intrinsic heterogeneity, respectively [24–26]. Here we propose a multistep chemometrical procedure that facilitates
260
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
the discrimination between pure fluids and their mixtures and the detection of minor mixture contributors. Using Raman spectroscopy, we were able to detect the contributions of blood in semen and of semen in blood stains at all the tested concentrations (5% blood in semen and 1% semen in blood stains are the smallest concentrations of the minority contributors used in our experiments). Smaller concentrations can be detected, but the probability of mixture detection at smaller concentrations is low. At the reported concentrations of 5% blood in semen and 1% semen in blood, only a part of spectra was distinguishable from the pure fluids. Our approach is based on the effective combination of regression and classification methods. Regression analysis assists with the selection of binary mixture Raman spectra that have distinct characteristics of both ingredients and can be easily distinguished from the spectra of pure body fluids. Discriminant Analysis based on the classes selected by regression significantly outperforms the direct application of regression or calibration methods alone. For the current study, calibration and classification were obtained using Support Vector Machines (SVM), a chemometrical method invented by Vapnik that can deal with ill-posed calibration problems and produce robust models in the case of spectral variations due to nonlinear interference [28]. SVM has already garnered a strong reputation in the characterization of a wide variety of objects, including biomaterials and forensic evidence [29–32]. The mixture of blood and semen was chosen because of the forensic community’s significant interest in its identification. If a victim of sexual assault cannot provide testimony, the ability to accurately reconstruct the crime scene based on the collected evidence is all the more important. Forensic investigators usually have to deal with body fluids and other types of evidence, which are considered important sources of DNA and can be used for serological and toxicological analysis. If a body fluid stain is detected, it is essential to know whether the stain comprises a mixture of fluids. Such a determination becomes more challenging if the crime was not reported for days or even years after the incident. The possibility of body fluid stain detection and identification at a crime scene will be greatly appreciated by the forensic community. There are several fluorescent tests already available for the detection of biological fluid stains; however, such tests can provide false-positive results when distinguishing between different types of body fluids [4–6], and forensic investigations require a comprehensive analysis that yields trusted results. Laurell immunoelectrophoretic separation of seminal and vaginal acid phosphatases (SAP and VAP) was among the first methods of semen and vaginal fluid mixture characterization [8]. The electrophoretic separation of the spermatozoa-specific lactate dehydrogenase (LDH) isoenzyme was used by Mokashi et al. to differentiate semen from blood and vaginal secretions [9]. Whitehead and co-workers described the technique based on amylase activity, presented as amylase-sensitive test-paper, which was able to detect saliva traces in blood or semen stains [10]. RNA and DNA profiling are the most common methods used for characterization of body fluid stains [1,11–23]. Much attention has been paid to the application of mRNA profiling for body fluid identification [17–23]. Fleming and Harbison recently developed a multiplex polymerase chain reaction (PCR) system, which can identify blood, saliva, semen and menstrual blood in individual stains or in mixtures of body fluids using messenger RNA (mRNA) [14]. Furthermore, they proposed vaginal-specific bacteria (Lactobacillus crispatus and Lactobacillus gasseri) as promising new markers for the forensic identification of vaginal secretions using a mRNA multiplex assay [15]. Despite recent progress, the development of non-destructive, easy-to-use, fast, and inexpensive methods for the identification of
body fluid composition is still anticipated by modern forensic science. We recently reported on the potential application of Raman spectroscopy to identify pure body fluid traces [33–40]. Other research groups have used Raman spectroscopy for analysis of fibers [41], drugs [42,43], and lipsticks [44], as well as ink [45], paint [45], explosives [46–49], bones [50], fingerprints [43,51,52] and condom lubricants [53]. Here we report the application of Raman spectroscopy in the analysis of blood/semen mixtures. The high selectivity and specificity of Raman spectroscopy to chemical and biochemical species coupled with advanced statistics allowed for the identification of pure blood and semen as well as their mixtures. The nondestructive character, high specificity and ability to extract information from small amounts of evidence are the main advantages of this method. The development of portable Raman instruments [54,55] may make it possible to bring these proposed methods directly to the crime scene [48]. 2. Materials and methods 2.1. Sample preparation and Raman microspectroscopy We have already reported the multidimensional Raman spectroscopic signatures of pure blood and semen [35,37]. If two body fluids are not thoroughly mixed and a dry sample contains small spots of pure fluids, the multidimensional Raman spectroscopic signatures can be used to identify the fluids. In the case of thoroughly mixed body fluids, a more complex approach may be required. Therefore, in this study, we focus only on samples prepared by mixing blood and semen thoroughly. Blood and semen samples were purchased from several companies including Bioreclamation, Inc., Lee Biosolutions, Inc., and Biological Specialty Corp. Blood/ semen stains were prepared using samples from two different anonymous individuals (a Caucasian male for semen and a Caucasian female for blood). Both donors were found to be negative for HbsAg, HCV, HIV-1&2, syphilis and HIV-1 antigen. All samples, in volumes of 10 mL each, were placed on microscope slides that were covered with aluminum foil to reduce fluorescence. Mixtures were prepared with different blood/semen ratios (5:95, 10:90, 20:80, 30:70, 40:60, 50:50, 70:30, 75:25, 85:15, 85.5:12.5, 92.75:6.25, 96.875:3.125 and 98.437:1.5625) by thoroughly shaking for 20 s. All samples were allowed to dry completely overnight. A Renishaw inVia confocal Raman spectrometer equipped with a research-grade Leica microscope and 50 long-range objective was used for the spectra acquisition. Raman spectra with 785-nm excitation were measured from 108 points using automatic mapping (Renishaw PRIOR automatic stage) from a sample area of 3.5 2.5 mm with a 10-s acquisition at each point. The laser power used on the dried samples was approximately 10 mW, and the spot size of the excitation beam was approximately 10 mm wide for the standard confocal mode. The spectral resolution was 0.8 cm 1, and the CCD camera was calibrated using a silicon standard. 2.2. Data treatment The Raman spectra obtained using the automatic mapping of all samples were first processed using the GRAMS/AI 7.01 software. After the cosmic ray interference removal, the Raman spectra were imported into MATLAB 7.4.0 for statistical analysis [56]. Normalization to the total area was performed to take into account the varying amount of background interference and the total offset variation. Because the visual inspection revealed the presence of a complex and varying fluorescent background, an adaptive iteratively reweighted penalized least squares (airPLS) algorithm [57] was used for baseline correction. The normalized and corrected spectra were subjected to dimension reduction by principal component analysis (PCA). The number of principal components was chosen based on significant factor analysis (SFA) and root-mean-square error of cross-validation (RMSECV) parameters of leave-one-out cross-validated PCA [58]. PCA scores were used for SVM regression by compiling the experimental data in to three groups: two groups were formed using the Raman spectra of the pure body fluids, and one group was formed using only the Raman spectra of the selected body fluid mixtures. Selection rules were established to minimize any interference of the third class with the first two (see Section 3 for details). The resulting three groups of data were used to build the SVM classification model. The full data set, including the Raman spectra of body fluids mixtures omitted in the previous step, revealed a high discrimination power of the developed SVM classification model.
3. Results and discussion Stains of body fluids mixtures are highly heterogeneous [39]. A single stain may have areas with variable composition ranging from practically pure semen to nearly pure blood. Fig. 1 shows selected raw Raman spectra acquired from pure blood and semen
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
261
Fig. 1. Selected characteristic raw Raman spectra acquired from pure blood (A) and semen (I) along with the spectra of blood/semen mixtures (B–H). Different colors are used to illustrate different spectral components. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
along with the spectra of blood/semen mixtures. Distinctive Raman bands of blood (754, 1003, 1226, and 1619 cm 1) [35] and semen (716, 830, 959, 1268, 1329, and 1671 cm 1) [37] can be used to identify their contributions in mixtures. Spots dominated by a single body fluid can be easily detected and identified using the corresponding multidimensional Raman signatures [35,37]. The Raman spectra of the mixtures are dominated by blood (Fig. 1). For example, when subject to visual inspection, all spectra with more than 50% blood do not have noticeable signs of semen contribution when a thoroughly mixed portion of the mixture is analyzed. Classical least squares (CLS), inverse least squares (ILS), partial least squares (PLS), principle component regression (PCR), leastsquares Support Vector Machines (LS-SVM) and artificial neural networks (ANN) are the most common multivariate methods applied to vibrational spectroscopic data [27,59,60]. However, the direct application of calibration methods to our experimental data
has a serious weakness. The heterogeneous character of body fluid stains significantly impairs the application of any calibration methods. Only part of the data is useful for calibration purposes; the rest of the data devaluate the calculations. 3.1. Weaknesses of the direct application of SVM classification and regression for the characterization of mixed traces of semen and blood SVM classification (one of the most widely used methods of biological sample classification) of 15 different semen/blood mixtures did not demonstrate proper discrimination (Fig. 2A). In Fig. 2A, each colored symbol corresponds to an experimental Raman spectrum; the color codes of the actual composition of the stains are as follows; class 1 (red symbols) and class 15 (black symbols) correspond to pure semen and blood stains, respectively, and classes 2–14 were assigned to body fluid mixtures starting
Fig. 2. SVM classification for Raman spectra of semen/blood mixtures. (A) All 15 mixtures with compositions varying from pure semen (red symbols) to pure blood (black symbols) were treated as separate classes. Each symbol corresponds to experimental Raman spectra. (B) A similar analysis performed with the same experimental data set, but with a different classification scheme: class 1 – pure semen, class 2 – mixtures, class – 3 pure blood. Both graphs indicate which predefined class the Raman spectra were assigned according to SVM. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
262
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
Fig. 3. (A) Contributions from blood in a semen/blood mixture calculated using Support Vector Machine (SVM) regression. Blue crosses and red triangles correspond to Raman spectra that were recorded from pure semen and blood samples, respectively. Different symbols correspond to different mixtures. Each symbol corresponds to a single Raman spectrum. (B) The selection of Raman spectra for the SVM Discriminant Analysis (DA) classification. The first and second groups include the Raman spectra of pure fluids only. Raman spectra of mixtures with significant contributions of both components comprise the third group. (C) Cross-validated results of the SVMDA classification. (D) SVMDA analysis of the experimental Raman spectra, including those omitted during the classification model development stage (B). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
from 5% blood in semen and finishing with 1% semen in blood stains (see Section 2.1). The y-coordinates of the colored symbols show to which predefined class the corresponding Raman spectrum was assigned according to SVM. Obviously the achieved assignments have to be refined, and some further investigation is needed. The ideal classification would provide a smooth graphical transition from pure semen (class 1) to pure blood (class 15); such a transition is not observed in the scattered plot seen in Fig. 2A. Many of the mixtures’ spectra were misclassified as pure semen; mixtures with a higher contribution of blood are not distinguishable from one another. Similar analysis performed with the same experimental data set, but with a different classification scheme (class 1 – pure semen, class 2 – mixtures, class – 3 pure blood), also demonstrated improper discrimination between fluids and their mixtures (Fig. 2B). In this case, about 40% of Raman spectra recorded from pure body fluids (red and blue symbols) was assigned to mixtures defined as class 2. SVM regression of the pretreated raw Raman data also demonstrates quite moderate results (Fig. 3A). Mixtures with low blood or semen content cannot be distinguished from the pure fluids, and the high variability of the data significantly impairs the assessment of the composition of unknown body fluid mixtures. 3.2. Conjugation of SVM classification and regression to facilitate the discrimination between pure fluids and their mixtures and the detection of minor mixture contributors To overcome the problem of discrimination between fluids and their mixtures, we combined classification and regression methods. Scheme 1 shows the main steps of the proposed approach. Pretreated Raman data were subjected to SVM regression. The results of SVM regression analysis were used to select Raman
spectra of binary mixtures that can be easily distinguished from the spectra of pure body fluids. In the next step, the Raman spectra of pure semen, pure blood and the Raman spectra selected in the previous step were assigned to three separate classes. As a result, blood/semen mixture spectra that demonstrated high similarity with the spectra of pure fluids were left out and not included in any class. This approach allows SVM to better identify the characteristic features of each class. The last main step of the procedure is the Discriminant Analysis (SVM classification) based on the defined classes. The SVM model, representation of the Raman spectra as points in virtual space with three distinct subspaces separated by clear gaps, was validated by different cross-validation techniques and tested with the previously omitted data. Therefore, in toto, the combination of regression and classification methods provides an appropriate quality of detection of the minor contributions from blood or semen in their mixtures. Fig. 3 demonstrates the results from the implementation of Scheme 1. A full, pretreated data set was used as the input file for the SVM calibration algorithm. Fig. 3 shows the results of calibration using leave-one-out cross-validation. The rough plot of the predicted concentration versus spectra reflects the level of spectral data variability after the performed filtration. We designated the sporadic spikes in the spectra recorded from spots as ‘‘non-characteristic’’ data for a given mixture composition. The high variability of the Raman spectra reflects the inherent heterogeneity of the samples. SVM regression demonstrates that mixtures with blood content higher than 80% cannot be distinguished from pure blood. On the other hand, 5% blood in a mixture resulted in a Raman spectrum that was distinguishable from the spectra of pure semen. Therefore, the lowest detectable levels of blood and semen according to the SVM regression model are approximately 5% and 25%, respectively. These observations were
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
263
Fig. 4. Selected results of SVMDA analysis of Raman spectra acquired from mixtures prepared with blood and semen samples from different donors. Colored symbols code different donors (pure body fluids) or different combinations of donors (mixtures). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Scheme 1. The main steps of the Raman data treatment procedure used for the identification of mixed traces of semen and blood.
used to group all of the experimental data into three groups (Fig. 3B). The first two groups were formed using the Raman spectra of pure body fluids, while the third group was built using all spectra falling within the 15–75% range. Boundaries were chosen to cut off the spectra of 100% pure fluids. Raman spectra of mixtures outside of the 15–75% range were not included in any group (Fig. 3B). The defined groups were assigned to classes 1 (semen), 2 (mixtures) and 3 (blood) and subjected to SVM Discriminant Analysis (Fig. 3C). The PCA method was used to reduce dimensionality and data block compression [58,60]. Cross-validation was performed via 1000 splits of the entire data set into training and test data sets. Each splitting was done by random selection of 75% and 25% of the data for training and test data sets, respectively. Fig. 3C demonstrates that 100% of the classified
Raman spectra were identified correctly. In the next step, the full data set was subjected to SVMDA analysis using the calculated SVMDA model (Fig. 3D). It is noticeable that most of the mixtures’ data were assigned as class 2 (mixtures), and only mixtures with 5% blood in the significant part of spectra were assigned to pure semen. These results considerably outperform both the direct classification without class selection and the composition determination using SVM regression methods used by themselves. The developed approach was additionally validated using Raman spectra acquired from mixtures prepared with blood and semen samples from six donors for each body fluid (different combinations were studied). Fig. 4 shows selected data obtained with new pure and mixed body fluids. The results of all validation studies were summarized in Table 1 which shows true positives, false positives, true negatives and false negatives classification rates for different blood contributions. These results demonstrate the feasibility of the approach. 3.3. Dry samples which were not thoroughly mixed The clear identification of pure body fluids and their mixtures is a promising result for the forensic community. However, there exists a small chance that the method will yield false positive and false negative identifications of pure body fluids and their mixtures (Table 1). We would like to emphasize that the case of not
Table 1 True positives, false positives, true negatives and false negatives classification rates of pure blood, pure semen and mixtures identification for different blood contributions. Blood contribution
0% 3% 5% 10% 20% 30% 40% 50% 70% 75% 85% 86% 93% 97% 99% 100%
Class 1: pure blood
Class 2: pure semen
Class 3: mixture
TP
FP
TN
FN
TP
FP
TN
FN
TP
FP
TN
FN
n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 1
0 0 0 0 0 0 0 0 0.01 0.01 0.01 0.01 0.03 0.21 0.64 n/a
1 1 1 1 1 1 1 1 0.99 0.99 0.99 0.99 0.97 0.79 0.36 n/a
n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0
1 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a
n/a 0.20 0.16 0.03 0.02 0 0 0 0 0 0 0 0 0 0 0
n/a 0.80 0.84 0.97 0.98 1 1 1 1 1 1 1 1 1 1 1
0 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a
n/a 0.80 0.84 0.97 0.98 1 1 1 0.99 0.99 0.99 0.99 0.97 0.79 0.36 n/a
0 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0
1 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 1
n/a 0.20 0.16 0.03 0.02 0 0 0 0.01 0.01 0.01 0.01 0.03 0.21 0.64 n/a
TP – true positives, FP – false positives, TN – true negatives and FN – false negatives, n/a – not applicable.
264
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265
are those of the authors and do not necessarily reflect those of the Department of Justice.
References
Fig. 5. Application of the developed SVMDA model (see text for details) to the experimental Raman spectra acquired from 1.8 2.3 mm area of dried samples of not thoroughly mixed blood and semen. Samples were prepared by random placing of blood and semen drops on aluminum foil. Spectra classified as blood, mixture and semen are presented by black, gray and white areas respectively.
thoroughly mixed body fluids will be easily distinguished through the identification of all pure body fluids involved (Fig. 5). In this paper, we provide solutions for the more complex problem of real mixtures. Scanning the entire stain could reveal even very small droplets of the intruding fluid, indicating spots to be subjected to further DNA analysis. Taking into account that those droplets are detectable even if the concentration of the minor contributor is small (5% or less), the overall effectiveness of Raman spectroscopy could be considerable. 4. Conclusions NIR Raman microspectroscopy was used to characterize stains of semen/blood mixtures. We have developed a new classification algorithm for the detection of pure body fluids and their mixtures. The classification methodology is based upon SVMDA of preprocessed multiple experimental Raman spectra collected during a scan of an unknown stain. The obtained results were tested by extensive cross-validation including validation with Raman spectra acquired from mixtures prepared with blood and semen samples from different donors. Studies of body fluid mixtures with low blood and semen concentrations indicated that the detection limit of a minor contributor was as low as a few percent. Further study is necessary to make the developed method applicable for real crime scene samples. Specifically, the effects of contamination, substrates, and aging should be investigated. Nevertheless, the results reported here clearly demonstrate the great potential of Raman spectroscopy in combination with advanced statistical analysis for the nondestructive and confirmatory identification of body fluid mixtures. The proposed methodology can be extended to the characterization of various complex mixtures, including but not limited to mixtures of other body fluids. Acknowledgements We are grateful to Dr. Barry Duceman, Director of Biological Sciences at the NY State Police Forensic Investigation Center for his valuable discussions and to Joseph De Rubertis for the manuscript preparation. This project was supported by Awards Nos. 2009-DN-BX-K196 and 2011-DN-BX-K551 awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice (I.K.L.). The opinions, findings, and conclusions or recommendations expressed in this publication
[1] T.A. Brettell, J.M. Butler, J.R. Almirall, Forensic science, Anal. Chem. 81 (12) (2009) 4695–4711. [2] R. Li, Forensic Biology, CRC Press, Boca Raton, FL, 2008. [3] J.M. Bogusz, Forensic Science, Elsevier Science, Amsterdam, The Netherlands, 2000, p. 2. [4] N. Vandenberg, R.A.H. van Oorschot, The use of polilights in the detection of seminal fluid, saliva, and bloodstains and comparison with conventional chemical-based screening tests, J. Forensic Sci. 51 (2) (2006) 361–370. [5] H.J. Kobus, E. Silenieks, J. Scharnberg, Improving the effectiveness of fluorescence for the detection of semen stains on fabrics, J. Forensic Sci. 47 (4) (2002) 819–823. [6] M. Stoilovic, Detection of semen and blood stains using polilight as a light source, Forensic Sci. Int. 51 (2) (1991) 289–296. [7] B.C. Pang, B.K. Cheung, Identification of human semenogelin in membrane strip test as an alternative method for the detection of semen, Forensic Sci. Int. 169 (1) (2007) 27–31. [8] P.J. Ablett, The identification of the precise conditions for seminal acid phosphatase (SAP) and vaginal acid phosphatase (VAP) separation by isoelectric focusing patterns, J. Forensic Sci. Soc. 23 (3) (1983) 255–256. [9] R.H. Mokashi, A.G. Malwankar, M.S. Madiwale, Detection of semen in the presence of blood and vaginal secretion, Indian Acad. Forensic Sci. 14 (2) (1975) 1–3. [10] A.E. Kipps, V.E. Quarmby, P.H. Whitehead, The detection of mixtures of blood and other body secretions in stains, J. Forensic Sci. Soc. 18 (3–4) (1978) 189–191. [11] C. Nussbaumer, E. Gharehbaghi-Schnell, I. Korschineck, Messenger RNA profiling: a novel method for body fluid identification by real-time PCR, Forensic Sci. Int. 157 (2–3) (2006) 181–186. [12] Y. Fujita, S. Kubo, Application of FTA technology to extraction of sperm DNA from mixed body fluids containing semen, Leg. Med. Tokyo 8 (1) (2006) 43–47. [13] M. Montesino, A. Salas, M. Crespillo, C. Albarran, A. Alonso, V. Alvarez-Iglesias, J.A. Cano, M. Carvalho, D. Corach, C. Cruz, A. Di Lonardo, R. Espinheira, M.J. Farfan, S. Filippini, J. Garcia-Hirschfeld, A. Hernandez, G. Lima, C.M. Lopez-Cubria, M. Lopez-Soto, S. Pagano, M. Paredes, M.F. Pinheiro, A.M. Rodriguez-Monge, A. Sala, S. Sonora, D.R. Sumita, M.C. Vide, M.R. Whittle, A. Zurita, L. Prieto, Analysis of body fluid mixtures by mtDNA sequencing: an inter-laboratory study of the GEP-ISFG working group, Forensic Sci. Int. 168 (1) (2007) 42–56. [14] R.I. Fleming, S. Harbison, The development of a mRNA multiplex RT-PCR assay for the definitive identification of body fluids, Forensic Sci. Int. Genet. 4 (4) (2010) 244–256. [15] R.I. Fleming, S. Harbison, The use of bacteria for the identification of vaginal secretions, Forensic Sci. Int. Genet. 4 (5) (2010) 311–315. [16] R. Li, Forensic Biology: Identification and DNA Analysis of Biological Evidence, CRC Press, Boca Raton, 2008. [17] C. Haas, B. Klesser, C. Maake, W. Bar, A. Kratzer, mRNA profiling for body fluid identification by reverse transcription endpoint PCR and realtime PCR, Forensic Sci. Int. Genet. 3 (2) (2009) 80–88. [18] M. Setzer, J. Juusola, J. Ballantyne, Recovery and stability of RNA in vaginal swabs and blood, semen, and saliva stains, J. Forensic Sci. 53 (2) (2008) 296–305. [19] J. Juusola, J. Ballantyne, Messenger RNA profiling: a prototype method to supplant conventional methods for body fluid identification, Forensic Sci. Int. 135 (2) (2003) 85–96. [20] J. Juusola, J. Ballantyne, mRNA profiling for body fluid identification by multiplex quantitative RT-PCR, J. Forensic Sci. 52 (6) (2007) 1252–1262. [21] J. Juusola, J. Ballantyne, Multiplex mRNA profiling for the identification of body fluids, Forensic Sci. Int. 152 (1) (2005) 1–12. [22] M. Alvarez, J. Juusola, J. Ballantyne, An mRNA and DNA co-isolation method for forensic casework samples, Anal. Biochem. 335 (2004) 289–298. [23] C. Haas, B. Klesser, A. Kratzer, W. Bar, mRNA profiling for body fluid identification, Forensic Sci. Int. Genet. 1 (2008) 37–38. [24] E. Smith, G. Dent, Modern Raman Spectroscopy – A Practical Approach, John Wiley & Sons Ltd., West Sussex, England, 2005. [25] J.M. Chalmers, P.R. Griffiths, Applications in Life Pharmaceutical and Natural Sciences, John Wiley & Sons, Ltd., Chichester, UK, 2002, p. 5. [26] R. Salzer, H.W. Siesler, Infrared and Raman Spectroscopic Imaging, vol. 510, WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim, 2009, p. 343. [27] V.A. Shashilov, I.K. Lednev, Advanced statistical and numerical methods for spectroscopic characterization of protein structural evolution, Chem. Rev. 110 (10) (2010) 5692–5713. [28] V. Vapnik, Statistical Learning Theory, John Wiley & Sons, New York, 1998. [29] L. Wang, Support Vector Machines: Theory and Applications, vol. 177, SpringerVerlag, New York, 2008, p. 448. [30] M. Tapiador, J. Go´mez, J.A. Sigu¨enza, Writer Identification Forensic System Based on Support Vector Machines with Connected Components, vol. 3029, SpringerVerlag, Ottawa, Canada, 2004, pp. 625–632. [31] B.M. Rodrigues, G. Peterson, K. Bauer, Fusion of steg analysis systems using Bayesian model averaging, in: I. Ray, S. Shenoi (Eds.), Advances in Digital Forensics IV, Springer-Verlag, New York, 2008. [32] B. Wang, X. Kong, X. You, Source camera identification using support vector machines, in: G. Peterson, S. Shenoi (Eds.), Advances in Digital Forensics V, Springer-Verlag, New York, 2009, p. 316.
V. Sikirzhytski et al. / Forensic Science International 222 (2012) 259–265 [33] K. Virkler, I.K. Lednev, Raman spectroscopy offers great potential for the nondestructive confirmatory identification of body fluids, Forensic Sci. Int. 181 (1–3) (2008) e1–e5. [34] K. Virkler, I.K. Lednev, Blood species identification for forensic purposes using Raman spectroscopy combined with advanced statistical analysis, Anal. Chem. 81 (18) (2009) 7773–7777. [35] K. Virkler, I.K. Lednev, Raman spectroscopic signature of blood and its potential application to forensic body fluid identification, Anal. Bioanal. Chem. 396 (1) (2010) 525–534. [36] K. Virkler, I.K. Lednev, Forensic body fluid identification: the Raman spectroscopic signature of saliva, Analyst 135 (3) (2010) 512–517. [37] K. Virkler, I.K. Lednev, Raman spectroscopic signature of semen and its potential application to forensic body fluid identification, Forensic Sci. Int. 193 (1–3) (2009) 56–62. [38] V. Sikirzhytski, A. Sikirzhytskaya, I.K. Lednev, Multidimensional Raman spectroscopic signatures as a tool for forensic identification of body fluid traces: a review, Appl. Spectrosc. 65 (11) (2011) 1223–1232. [39] V. Sikirzhytski, K. Virkler, I.K. Lednev, Discriminant analysis of Raman spectra for body fluid identification for forensic purposes, Sensors 10 (4) (2010) 2869–2884. [40] A. Sikirzhytskaya, V. Sikirzhytski, I.K. Lednev, Raman spectroscopic signature of vaginal fluid and its potential application to forensic body fluid identification, Forensic Sci. Int. 216 (1–3) (2012) 44–48. [41] J. Thomas, P. Buzzini, G. Massonnet, B. Reedy, C. Roux, Raman spectroscopy and the forensic analysis of black/grey and blue cotton fibres. Part 1: investigation of the effects of varying laser wavelength, Forensic Sci. Int. 152 (2–3) (2005) 189–197. [42] C. Hodges, J. Akhavan, The use of Fourier transform Raman spectroscopy in the forensic identification of illicit drugs and explosives, J. Mol. Spectrosc. 46 (1990) 303–307. [43] M.J. West, M.J. Went, The spectroscopic detection of drugs of abuse in fingerprints after development with powders and recovery with adhesive lifters, Spectrochim. Acta A: Mol. Biomol. Spectrosc. 71 (5) (2009) 1984–1988. [44] D. Broughton, C. Rodger, The in situ analysis of lipsticks by surface enhanced resonance Raman scattering, Analyst (1998) 1823–1826. [45] W.D. Mazzella, P. Buzzini, Raman spectroscopy of blue gel pen inks, Forensic Sci. Int. 152 (2-3) (2005) 241–247. [46] E.M. Ali, H.G. Edwards, I.J. Scowen, In situ detection of single particles of explosive on clothing with confocal Raman microscopy, Talanta 78 (3) (2009) 1201–1203.
265
[47] R.M. Burks, D.S. Hage, Current trends in the detection of peroxide-based explosives, Anal. Bioanal. Chem. 395 (2) (2009) 301–313. [48] D.S. Moore, R.J. Scharff, Portable Raman explosives detection, Anal. Bioanal. Chem. 393 (6-7) (2009) 1571–1578. [49] S.K. Sharma, A.K. Misra, B. Sharma, Portable remote Raman system for monitoring hydrocarbon, gas hydrates and explosives in the environment, Spectrochim. Acta A: Mol. Biomol. Spectrosc. 61 (10) (2005) 2404–2412. [50] G. McLaughlin, I. Lednev, Potential application of Raman spectroscopy for determining burial duration of skeletal remains, Anal. Bioanal. Chem. 401 (8) (2011) 2511–2518. [51] R.M. Connatser, S.M. Prokes, O.J. Glembocki, R.L. Schuler, C.W. Gardner, S.A. Lewis Sr., L.A. Lewis, Toward surface-enhanced Raman imaging of latent fingerprints, J. Forensic Sci. 55 (6) (2010) 1462–1470. [52] E. Widjaja, Latent fingerprints analysis using tape-lift, Raman microscopy, and multivariate data analysis methods, Analyst 134 (4) (2009) 769–775. [53] T. Coyle, N. Anwar, A novel approach to condom lubricant analysis: in situ analysis of swabs by FT-Raman spectroscopy and its effects on DNA analysis, Sci. Justice 49 (1) (2009) 32–40. [54] A. Eckenrode, E.G. Bartick, S. Harvey, M.E. Vucelick, B.W. Wright, R.A. Huff, Portable Raman spectroscopy systems for field analysis, Forensic Sci. Commun. 3 (2001). [55] F. Yan, T. Vo-Dinh, Surface-enhanced Raman scattering detection of chemical and biological agents using a portable Raman integrated tunable sensor, Sens. Actuators B (2007) 61–66. [56] B.M. Wise, N.B. Gallagher, R. Bro, J.M. Shaver, W. Windig, J.S. Koch, PLS Toolbox 3.5 for Use with Matlab, vol. 17, Eigenvector Research Inc., Manson, 2005. [57] Z.M. Zhang, S. Chen, Y.Z. Liang, Baseline correction using adaptive iteratively reweighted penalized least squares, Analyst 135 (5) (2010) 1138–1146. [58] E.R. Malinowski, Factor Analysis in Chemistry, vol. 3, John Wiley & Sons Inc., New York, 2002. [59] J.E. Franke, Inverse least squares and classical least squares methods for quantitative vibrational spectroscopy, in: J.M. Chalmers, P.R. Griffiths (Eds.), Handbook of Vibrational Spectroscopy, John Whiley & Sons, Ltd., New York, 2001, pp. 2276– 2292. [60] T. Hasegawa, Principle component regression and partial least squares modeling, in: J.M. Chalmers, P.R. Griffiths (Eds.), Handbook of Vibrational Spectroscopy, John Whiley & Sons, Ltd., New York, 2001, pp. 2293–2312.