Laboratory-made Kishk from wheat, oat and barley: 2. Compositional quality and sensory properties

Food Research International, Vol. 30, No. 5, pp. 319±326, 1997 # 1998 Canadian Institute of Food Science and Technology Published by Elsevier Science Ltd Printed in Great Britain PII: S0963-9969(97)00055-0 0963-9969/98 $19.00+0.00

Laboratory-made Kishk from wheat, oat and barley: 2. Compositional quality and sensory properties A. Y. Tamime,a* D. D. Muir,b M. N. I. Barclay,a M. Khaskhelia & D. McNultyc a

SAC Auchincruive, Food Science and Technology Department, Ayr, KA6 5HW, UK b Hannah Research Institute, Ayr, KA6 5HL, UK c Biomathematics & Statistics Scotland (BioSS), University of Edinburgh, Edinburgh, EH9 3JZ, UK Kishk was made from a dough containing low-fat yoghurt, parboiled `cracked' cereal (Burghol) and salt. The dough was then dried and ground to a ¯our. The composition and sensory properties of Kishk made from wheat, oat and barley Burghol were investigated. The chemical analysis (g 100 gÿ1 on dry matter basis) of the Kishk fell within the following ranges: protein 18.2±20.6, fat 6.4±10.7 and carbohydrates 62.0±68.6. The moisture content averaged 8.4%, and the ®bre and -glucan contents were highest in Kishk made with barley and oat Burghol, respectively. Lactic acid was the major organic acid present in the products. The mono-unsaturated fatty acid content (34%) was the highest in oat Kishk. Appreciable quantities of Fe, Cu and Mn were found in all the Kishks paralleling the mineral composition of the cereal of which it was made. The sensory pro®les of a hot porridge-like gruel of nine samples of Kishk showed substantial di€erences between these products made with di€erent cereals. Di€erences in mouthfeel (grainy, sticky and slimy character) were associated with cereal type. Partial least squares regression (PLS2) models derived from the chemical composition were successfully ®tted, after cross-validation, for grainy, sticky and slimy character. Only the model for grainy character was of predictive value. # 1998 Canadian Institute of Food Science and Technology. Published by Elsevier Science Ltd Keywords: Kishk, composition, odour, ¯avour, after-taste, mouth-feel, Partial Least Squares Regression.

INTRODUCTION

1998a,b; Tamime and McNulty, 1998), the sensory pro®les of 15 varieties of commercial Kishk have been studied (Muir et al., 1995), and clear di€erence was found between the sensory character of Kishk made from wheat Burghol and yoghurt made from either bovine or caprine milk. However, little is known on the contribution made by the cereal component of Kishk to overall quality. In a previous paper, we have described the preparation of Burghol from wheat, oats and barley and identi®ed the di€erences between the products (Tamime et al., 1997). The sensory properties of nine di€erent types of Kishk made from these di€erent kinds of Burghol have been studied, and the prediction of sensory character from measurements of composition has been investigated.

Many `new' products are based on traditional food which has hitherto received limited exposure within ethnic or cultural groups. However, before such traditional products are introduced to new markets, it is prudent to characterise the existing product range. This approach has been applied to KishkÐa dried mixture of fermented milk and cereal (parboiled `cracked' wheat known as Burghol), widely consumed in the region between the eastern Mediterranean and the Indian subcontinent (Tamime and O'Connor, 1995). In addition to chemical and microbial quality (Tamime et al., *To whom correspondence should be addressed. Fax: 0044 1292 525071; e-mail: [email protected] 319

320

A. Y. Tamime et al.

MATERIALS AND METHODS Materials Four varieties each of barley and oat and one variety of wheat (i.e. coarse) Burghol were used in the preparation of Kishk (for details see Tamime et al., 1997). Skimmed milk powder (SMP) (Dairy Crest Ingredients, Surrey, UK), anhydrous milk fat (AMF) (Bodfari Producers Ltd., Chester, UK), and freeze-dried yoghurt starter culture MY 087 (RhoÃne-Poulenc Texel (UK) Ltd., Cheshire, UK) were used for the production of low-fat yoghurt. All chemicals used for analysis were obtained from BDH Chemicals Ltd., Poole, UK (analytical grade). Preparation of Kishk Production of yoghurt SMP was reconstituted to ca 11 g 100 gÿ1 total solids at 40 C, warmed to 60 C and melted AMF (65 C; ca 1.5 g 100 gÿ1) was added using a hand whisk to form a coarse emulsion (Barrantes et al., 1994). This base was homogenised (17.3 MPa; Rannie Machine Works Ltd., DK-2620 Albertslund, Denmark), heated to 90 C for 10 min then cooled to 45 C in a water bath. Direct-tovat culture was inoculated (19 g 100 lÿ1) and fermentation was carried out at 42 C for ca 4 h or to until the pH fell to 4.6 at which time the fermentate was transferred to a cold store. The initial counts in the freezedried starter culture of Streptococcus themophilus and Lactobacillus delbrueckii sp. bulgaricus were 2.9  1011 and 6.8  108 cfu gÿ1, respectively. Kishk-making Each type of Burghol was washed with water (ca 50 C) to remove residual bran or ¯our dust, then air dried. The Burghol was then mixed with fermentate in a ratio of 1:4. Salt was added to the fermented milk/cereal mixture (0.85 g 100 gÿ1) and the resulting dough was stored for six days at ca 20 C. On each of the six days, the dough was thoroughly kneaded to encourage secondary fermentation. At completion of this process, the dough was formed into nuggets (3±5 cm in diameter), placed into stainless steel trays and dried in a bakery oven (50 C for 8±9 h; Double D Food Engineering, West Lothian, UK). The dried nuggets were milled to a powder using a hammer mill (Christy Noris Ltd., Chelmsford, UK; mesh size 0.8 mm). The dried materials were stored in airtight containers and kept in the dark, under refrigeration, until tested. Proximate composition Protein, fat, ash, moisture, ®bre, free sugar, -glucan and phytic acid contents of di€erent types of Kishk were determined using the methods described by

Tamime et al. (1998a,b). The total carbohydrates content was calculated by di€erence [total solids±(protein, + fat + ash)]. The crude protein content was calculated from the total nitrogen using a factor based on the relative ratios of milk and cereal protein: protein ˆ nitrogen  ‰…4=5  6:38† ‡ 1=5  5:7†Š nitrogen  6:244

Mineral content The mineral contents (Ca, P, Mg, Zn, Fe, Cn, Mn, K and Na) of the Kishk samples were measured according to MAFF (1986), as described in Tamime et al. (1997). Organic acid composition Individual organic acids (orotic, citric, pyruvic, lactic, uric/formic, acetic and hippuric) in the Kishk samples were determined by a modi®cation of the method of Marsili et al. (1981) in which the concentration of H2SO4 was halved. This change resulted in resolution of orotic and citric acids during chromatography without a concomitant e€ect on the retention times of the other organic acids (Tamime et al., 1998a). Fatty acid analysis The fatty acids content in Kishk was determined by gas liquid chromatography (GLC), on lipid extracted by the Rose-Gottlieb method, according to BSI (1992) as described by Barrantes et al. (1994). A set of fatty acid standards, dissolved in heptane (Sigma-Aldrich Company Ltd., Poole, Dorset, UK) was used to externally calibrate the chromatograph. Response factors were automatically determined by the data processing system (Chromperfect; Justice Innovations Chromatography Data Systems, Mountain View, California, USA). Lipid (50 mg) extracted from Kishk was weighed into stoppered test tubes and dissolved in 2 ml hexane. Later, 0.2 ml of methanolic KOH (2%) were added and shaken until the solution became clear. The glycerol fraction settled to the bottom of the test tube after standing for ca 10 min. Sensory pro®ling Vocabulary Sensory pro®ling was carried out using a protocol successfully applied to a diverse range of commercial samples (Muir et al., 1995). During the preliminary stage of that study the panel of assessors and the sensory scientists agreed on a list of descriptors by which to characterize odour, ¯avour, aftertaste and mouth-feel. The panel, after acclimatization to the sensory properties of

Laboratory-made Kishk from wheat, oat and barley: 2.

321

commercial Kishk, used this vocabulary to pro®le a complete set of samples. The attribute ratings were analysed and redundant or poorly understood terms were deleted from, and additional descriptors inserted into, the experimental vocabulary. The ®nal vocabulary, derived for commercial products (Muir et al., 1995), was applied without modi®cation to the laboratoryprepared Kishk in the present study. The pro®le comprised seven attributes to describe aroma (overall intensity, creamy/milky, acid/vinegary/sharp, fruity/ sweet, cooked, cereal, cardboard), 10 ¯avour attributes (overall intensity, creamy/milky, acid/vinegary/sharp, fruity/sweet, cooked, cereal, cardboard, apple, bitter, salty), ®ve descriptors of aftertaste (overall intensity, persistence, acid/vinegary/sharp, cereal, cardboard), and ®ve terms used to describe mouth feel (viscosity, grainy/ ¯oury/chalky texture, sticky/gluey texture, slimy texture, mouth-coating character).

Minitab Ltd., Coventry, UK). In previous sensory studies, order of tasting e€ects have been noted (Muir and Hunter, 1991/2). Therefore, a statistical model was ®tted with e€ects for sample, assessor, order of tasting and replicate. Signi®cant sample e€ects (F-test; p  0.05) are reported as panel mean values. The prediction of the ratings of the sensory attributes (Y-matrix) from chemical composition (X-matrix) was investigated by Partial Least Squares Regression (PLS2, Unscrambler v.5.03, Camo A/S, Trondheim, Norway). The method of deriving the optimum PLS2 model was based on the principles proposed by Martens and Nñs (1989). Both the X- and Y-matrix values were pre-processed by the normalisation routine in Unscrambler. This method of scaling replaces every element in the matrix according to the formula:

Assessors The panel of assessors was an external panel of 15 female non-smokers, aged 39±55, whose sole duties are sensory characterisation. The panel were selected on the basis of sensory acuity and consistency and are highly experienced in pro®ling a wide range of foods. Panel performance is assessed by the principles described by Hunter et al. (1995).

where Mavg(i) = [M(i,1)+M(i,2) + ...+M(i,k)+ ...M (i,K)]/K Validation of predictive models was facilitated by cross-validation (nine segments). The optimum number of PLS Factors was selected by choosing the ®rst local minimum in the residual Y-variance plot. Variances explained by the optimum PLS2 model are reported after calibration and after cross-validation. -coecients for each Y-variate in the optimum PLS2 model are also reported.

Sample preparation Kishk samples made from di€erent cereal bases were treated in an identical fashion. Each dried powder was mixed with four parts cold tap water and brought to boiling point with continuous stirring. After simmering for ®ve min with occasional stirring, the product was immediately served hot in ceramic cereal bowls. Sample presentation Sample assessment was carried out in isolated air-conditioned booths with controlled lighting. Assessors cleansed their palate with a plain, water biscuit and some cold water then rated each sample for the attributes listed above. Two samples were presented in each tasting session, and several sessions (4±6) were held on each day. The order of presentation of samples was balanced (Muir and Hunter, 1991/2). Rating of attributes was on an undi€erentiated scale with anchor points (absent, extremely strong) and was facilitated by an interactive, computer-assisted data collection system (Williams et al., 1996). Samples were pro®led twice. Data analysis Mean sample e€ects were computed by balanced analysis of variance (Genstat 5.3; copyright 1990 Lawes, Agricultural Trust, Rothamsted Experimental Station, UK) or, in the case of sensory ratings by applying the General Linear Model (Minitab 10;

Mnew …i; k† ˆ M…i; k†=Mavg…i†

RESULTS AND DISCUSSION Gross chemical composition The average chemical composition of nine samples of Kishk is shown in Table 1. Although there was no statistically signi®cant variation in the moisture content (i.e. 7.8 to 8.8 g 100 gÿ1), the data was converted to a dry matter basis (DMB) for comparative purposes (Tamime et al., 1998a). Thus, the protein, carbohydrate, fat and ash contents (g 100 gÿ1) ranged between 18.2 and 20.6, 62.0 and 68.6, 6.4 and 10.7, and 6.4 and 6.8, respectivelyÐi.e. within the range of commercial product (Tamime et al., 1998a). The greatest variation between cereal types was found in the fat and carbohydrate constituents. Oat Kishk had the highest fat content (mean 9.7 g 100 gÿ1), a re¯ection of the higher fat content of oat Burghol (Tamime et al., 1997). Salt content was within the acceptable range for Kishk made in the Lebanon, although lower than that found in Egypt (up to 10.8%; Tamime and O'Connor, 1995).The gross chemical composition of the nine di€erent types of Kishk directly re¯ected the composition of Burghol used in its preparation and con®rm that the laboratory-prepared material is similar to the commercial product.

322

A. Y. Tamime et al. Table 1. Chemical composition of Kishk made from Burghol manufactured from oats, barley and wheat Concentration by cereal type and cultivara Oat

Component Proximal composition Moisture Crude protein Total carbohydrate Starch Soluble sugars Fibre -glucan Phytic acid Total fat Fatty acidsd C4:0±C18:0 C18:1 C18:2 C18:3 Organic acidse Orotic Citric Pyruvic Lacticf Uric/formic Acetic Hippuric Mineralsg Sodium Potassium Phosphorous Calcium Magnesium Copper Zinc Iron Manganese

Adamo

Dula

Barley

Matra

Wheat

Valient Camargue Maghee Marinka Pastoral

Salabi Sign.b SEDc

8.12 20.52 62.02 39.03 7.92 6.60 2.20 0.67 10.71

8.02 20.59 63.22 39.10 9.18 7.14 2.53 0.70 9.46

8.22 20.61 63.43 40.72 9.82 6.92 2.10 0.83 9.28

8.38 20.36 63.47 42.37 8.75 6.15 1.89 0.69 9.51

7.75 18.17 68.23 45.15 7.43 7.35 1.73 0.49 7.05

8.84 18.16 68.64 43.54 7.08 9.29 1.57 0.51 6.62

8.54 19.57 67.21 43.68 7.75 8.97 2.22 0.45 6.81

8.41 19.18 67.68 44.62 8.12 7.88 1.81 0.44 6.64

8.86 20.32 66.74 40.84 9.83 8.98 1.10 0.48 6.42

Ð *** *** *** ** *** *** *** ***

0.83 0.30 0.68 0.76 0.62 0.62 0.19 0.06 0.42

45.71 37.27 15.04 1.99

51.34 33.66 12.93 2.08

52.90 32.64 12.48 1.99

52.37 32.99 12.86 1.78

61.45 24.62 11.46 2.46

61.51 25.79 10.61 2.09

63.88 24.85 9.80 1.47

63.13 24.62 10.12 2.12

58.19 25.82 13.10 2.89

*** ** * Ð

1.10 1.02 1.27 0.44

Ð Ð Ð Ð * Ð Ð

3.19 35.23 9.19 0.69 2.04 10.38 3.92

Ð Ð *** *** *** *** *** *** ***

45.40 25.90 20.10 13.50 3.70 0.06 0.17 3.19 0.17

21.5 220 43.5 19.3 13.5 44.0 34.5

22.0 204 42.0 19.8 12.0 48.0 43.5

23.5 245 40.5 20.2 12.0 42.5 40.0

22.5 204 44.5 20.0 12.0 63.5 39.0

1348 1339 1358 1395 760 761 741 7460 611 638 644 623 436 462 485 455 122 137 137 118 0.18 0.19 0.29 0.34 3.54 3.81 4.14 3.99 7.57 10.37 10.06 9.20 2.53 3.30 3.37 3.42

24.5 272 67.5 18.2 16.5 50.0 32.5

23.0 168 51.0 18.5 18.0 50.5 34.0

23.0 231 52.5 19.4 17.5 55.5 38.0

1399 754 571 439 111 0.31 2.96 8.99 0.90

1374 740 554 437 110 0.29 3.65 7.88 0.82

1309 730 542 419 101 0.25 3.45 6.53 1.02

24.0 255 46.5 18.4 20.5 75.5 37.5

18.5 181 52.0 19.2 23.5 60.0 33.5

1341 1360 789 799 527 552 406 439 104 116 0.36 0.42 2.86 3.60 7.12 21.58 0.71 1.72

a

Composition g 100ÿ1; computed on a dry matter basis; results are average of replicate determinations on material from three trials. Statistical signi®cance of cultivar e€ect; Ð = not signi®cant; * = p  0.05; **p =  0.01; *** = p  0.001; n = 27. c SED = standard error of di€erence. d Fatty acids expressed as percentage of total (m mÿ1) fatty acid. e Organic acidsÐorotic to hippuricÐexpressed as g gÿ1 Kishk, except f expressed as mg gÿ1. g MineralsÐsodium to manganese expressed as mg 100 gÿ1; computed on dry matter basis. b

Carbohydrates The Kishk samples contained substantial and variable amounts of starch (digestible and non-digestible fractions), soluble sugars, ®bre, -glucan and phytic acids (Table 1). The average starch content of Kishk made from oat and wheat Kishk was lower (40.4 g 100 gÿ1) than that of Kishk made from barley (44.2 g 100 gÿ1). The opposite pattern was observed for the soluble sugar content. The mean phytic acid contents of Kishk made from oat, barley and wheat were 0.72, 0.48 and 0.48 g 100 gÿ1, respectivelyÐlower than that found in commercial Lebanese samples (0.96; Tamime et al., 1998b). Because phytate can prevent the absorption of dietary calcium, iron and zinc (Passmore

and Eastwood, 1986), the choice of an appropriate cereal variety can reduce such undesirable dietary e€ects. The average ®bre content of Kishk made from oats, barley and wheat was 6.70, 8.37 and 8.98 g 100 gÿ1, respectively. These di€erences re¯ect cereal type and the eciency of de-husking and the Burghol-making (Tamime et al., 1997). The high levels of -glucan (>2 g 100 gÿ1) found in Kishk made from oats (Adamo, Dula and Matra) and one barley sample (Marinka) may represent a potential dietary bene®t. Organic acids The concentration (g gÿ1) of organic acids in the Kishk samples made with oat, barley or wheat Burghol

Laboratory-made Kishk from wheat, oat and barley: 2. is shown in Table 1. No appreciable di€erences of organic acids pro®les in di€erent Kishk samples were observed suggesting that the type of cereal used had no e€ect on the metabolic activity of Lb. delbrueckii sp. bulgaricus and S. thermophilus. The level of lactic acid in the experimental Kishk was 1.7-fold lower than that of commercial samples of Lebanese Kishk (Tamime et al., 1998a). This implies that control of the secondary fermentation during commercial production (i.e. when the yoghurt and Burghol were mixed and left at ambient temperature for a week) was poor. In addition, the acetic acid content of the laboratory material was 50% lower than in commercial Kishk. Furthermore, propionic acid was not detected in Kishk made with either oat, barley or wheat Burghol. These contrasts between laboratory and commercial product are the consequence of the use of di€erent types of starter culture. Nevertheless, the combined levels of organic acids in the experimental Kishk samples were suciently high to ensure the microbiological safety of the product. Fatty acids The fatty acid pro®les of Kishk made from di€erent types of cereal are shown in Table 1. The variations arise from the cereal lipid because a single source of milk fat was used for preparation of yoghurt. In general, cereal lipid does not contain appreciable amounts of short-chain [(C4:0±C14:0 see Table 1] fatty acid, but there is variation between cultivars in the distribution of long-chain fatty acid (Paul et al., 1980; Pomeranz, 1991). Oat-based Kishk was substantially richer in oleic (C18:1) and linoleic acid (C18:2) than product made from barley. Minerals The proximate concentrations of minerals in 9 samples of Kishk are also shown in Table 1. To aid comparison of the di€erence between Kishk made from di€erent cereal types, star plots were constructed (Fig. 1). Each attribute is assigned a vector, the length of which is determined by the magnitude of the attribute, expressed as a proportion of the maximum value within the

Fig. 1. Star plots for average mineral content of Kishk made from di€erent cereals. Each vector represents a di€erent mineral and the length of each vector is proportional to the maximum concentration of that element within the sample set.

323

sample set. The major variations in mineral content were for manganese, iron and copper (Fig. 1). Sensory properties The discriminant attributes, i.e. those with signi®cant sample e€ects ( p  0.05), are shown in Table 2. Only 14 of the original 27 attributes were useful discriminants. This ®nding re¯ected the fact that the Kishk was manufactured under closely controlled laboratory conditions and that a common yoghurt base was used throughout. Although this reduction in diversity was noted during acclimatisation trials, it was resolved to use the full vocabulary to ensure comparability between the present and earlier studies (Muir et al., 1995). All samples were assigned high ratings for cooked and cereal odour whilst creamy/milky character was of secondary importance (Table 2). Acid, salty and cardboard ¯avours were perceived as important ¯avour notes with apple being less important. The Kishk had an intense and persistent aftertaste. As might be anticipated, the hot gruels made from Kishk were also perceived as viscous, grainy, sticky and slimy (Table 2). There was no evidence of systematic di€erences in odour, ¯avour or after-taste being associated with cereal type. However, mouth-feel was substantially di€erent. Products based on oats and barley were more viscous, sticky and slimy and less grainy than Kishk made from wheat. Prediction of sensory character from chemical composition Prediction of the individual sensory attributes from chemical composition was investigated using PLS2. The chemical attributes attaining statistical signi®cance (Table 1), excepting total carbohydrate because it was a derived value, formed the X-matrix (9  18) and the sensory attributes detailed in Table 2 formed the Y-matrix (9  14). No satisfactory prediction could be made for aroma, ¯avour or after-taste. However, models were ®tted for grainy, sticky and slimy mouth-feel (Table 3). After calibration the Y-variance explained ranged from 62.5±86.7%. However, after cross-validation, the model was found to be unsatisfactory because less than 45% of the Y-variance was modelled (Table 3). Further simpli®cation was achieved by using only the four attributes describing mouth- feel to form the Ymatrix (9  4) and excluding the mineral information, after it was found to make little contribution, from the X-matrix (9  11). The new model explained substantially more of the variance in the sensory attributes (Table 3). In particular, grainy character was well predicted. The sample scores plot and X- and Y-loadings plots for the optimum (2 Factor) model are shown in Fig. 2(a)±(c). The samples representing di€erent cereal types were very clearly separated by the scores on PLS Factors 1 and 2 [Fig. 2(a)]. Examination of the X-loadings plot

324

A. Y. Tamime et al.

Table 2. Panel mean ratings for signi®cant (F test; P  0.05) sensory attributes of Kishk made from Burghol manufactured from oats, barley and wheat Panel mean rating by cereal type and cultivara Oat Attribute Aroma Creamy Cooked Cereal Flavour Overall intensity Acid Cardboard Apple Salty After-taste Overall intensity Persistence Mouthfeel Viscosity Grainy Sticky Slimy

Barley

Wheat

Adamo

Dula

Matra

Valient

Camargue

Maghee

Marinka

Pastoral

Salibi

SEDb

13.6 53.2 46.2

16.3 51.3 37.4

13.4 47.1 41.4

16.6 49.5 41.5

11.9 49.0 40.5

12.1 49.5 44.4

12.1 47.2 36.9

13.1 45.4 41.1

15.9 46.4 43.0

2.9 3.5 4.6

63.6 55.0 35.7 18.6 36.8

59.7 57.5 38.2 19.4 23.6

61.9 49.2 34.5 23.5 31.1

65.1 55.5 36.1 19.8 30.7

57.4 46.9 28.5 21.7 28.9

62.5 49.3 41.3 16.1 33.2

62.5 51.1 36.1 22.0 36.0

61.7 53.6 37.2 19.1 30.0

61.9 51.1 26.4 23.9 32.0

3.0 5.7 5.6 4.4 6.0

55.2 51.6

55.9 53.7

54.3 51.4

55.3 52.5

47.5 46.3

55.8 52.9

53.4 51.4

51.6 49.2

53.9 50.3

3.1 3.5

71.8 39.7 62.1 22.5

70.5 29.1 62.5 29.9

71.6 33.0 63.4 30.3

70.0 29.0 62.5 29.5

68.5 33.1 53.3 26.0

71.9 41.6 58.4 25.7

69.2 43.5 60.0 21.2

73.4 41.5 60.9 17.2

64.7 48.4 43.2 16.1

2.8 4.6 4.4 4.5

a

Scale 0±100; values are based on duplicate assessments by 15 trained assessors. SED = standard error of di€erence.

b

Table 3. Prediction of sensory attributes from chemical composition by partial least squares regression (PLS2). Variance explained by models after calibration and after cross-validation PLS2 modela

Sensory attribute All attributes

Mouth feel Grainy Sticky Slimy

Mouth-feel attributes only

Variance explained %b

No. factors

Variance explained %b

No. factors

78.6 (43.8) 62.5 (nilc) 86.7 (11.8)

4 (1) 4 (1) 4 (1)

88.9 (83.3) 66.7 (10.0) 63.6 (30.8)

2 (2) 2 (2) 2 (2)

a

X- and Y-variates normalised. Y-variance explained after calibration; Y-variance explained after cross-validation in parenthesis. c Optimum model based on factor coincident with minimum in residual variance after cross-validation (nine segments), nil indicates no satisfactory model could be ®tted. b

Fig. 2. PLS2 regression for prediction of sensory attributes from chemical composition of Kishk. (a) Scores plot on Factors 1 and 2 for Kishk made from cultivars of oat (Adamo +, Dula *, Matra x, Valient #), barley (Camargue ^, Maghee ~, Marinka !, Pastoral &) and wheat (Salibi *); (b) vector loadings plot for chemical attributes on PLS Factors 1 and 2; (c) Vector loadings plot for sensory attributes describing mouth-feel on PLS Factors 1 and 2.

Laboratory-made Kishk from wheat, oat and barley: 2.

325

Fig. 3. PLS2 regression for prediction of sensory attributes from chemical composition of Kishk. -coecients for 2-factor model for attributes describing mouth-feel.

[Fig. 2(b)] revealed that the compositional attributes making most contribution to Factors 1 and 2 were fatty acid composition, uric/formic acid and starch concentration. As expected from the variance explained, grainy character dominated the Y-loadings plot [Fig. 2(c)]. The inverse relation between grainy character and sticky/ slimy mouth-feel is also clearly illustrated. The -coecient from the prediction models are shown in Fig. 3. Although the e€ect of starch content and fatty acid type can be rationalised in terms of the swelling behaviour of starch and its consequential e€ect on mouth-feel, the apparently important in¯uence of uric/formic acid is less easily explained. It serves as a reminder that soft modelling does not imply causality. It is probably that uric/ formic acid content is associated with some other property of the product which has not been modelledÐe.g. di€erences in gelatinization behaviour of the starch. CONCLUSIONS Variations in the proximate composition, carbohydratesbased nutrients, fatty acids and mineral contents in Kishk made with di€erent cereal Burghol were observed. Kishk is a good dietary source of -glucan, ®bre, mono-

unsaturated fatty acids and certain minerals especially when made from oat Burghol. The high level of iron in wheat Kishk should not be overlooked. Substantial differences in sensory character were noted between the Kishk made with di€erent cereals. In particular, di€erences in mouth-feel (grainy, sticky and slimy character) were associated with cereal type. The prediction of sensory character from the chemical composition was investigated by partial least squares regression (PLS2) and models were successfully ®tted, after cross-validation, for grainy, sticky and slimy character. Nevertheless, only the model for grainy character explained a suciently high proportion of the variance (83%) to have practical utility. ACKNOWLEDGEMENTS The authors thank Mrs I. Hamilton, Miss A. Mair, Mr T. McCreath, Mrs C. Shankland and Mrs K. Smith for skilled technical assistance. SAC, HRI and BioSS receive ®nancial support from the Scottish Oce of Agriculture, Environment and Fisheries Department (SOAEFD) and Mr M. Khaskheli is indebted to The World Bank for ®nancial support.

326

A. Y. Tamime et al.

REFERENCES Barrantes, E., Tamime, A. Y., Muir, D. D. and Sword, A. M. (1994) The e€ect of substitution of fat by microparticulate whey protein on the quality of set-type, natural yogurt. J. Soc. Dairy Technol. 47, 61±68. BSI (1992) Fats and Fatty Oils. BS: 684 Section 2. 34, British Standards Institution, London. Hunter, E. A., Muir, D. D. and Brennan, R. M. (1995) Comparison of the performance of an external panel with an internal panel. In Proceedings 4eÁmes Journees Europeennes Agro-Industrie et Methods Statistiques, 7±8 December 1995, pp. 77±85. Dijon. MAFF (1986) The Analysis of Agricultural Materials. Reference No. 427, HMSO, London. Marsili, R. T., Ostapenko, H., Simmons, R. E. and Green, D. E. (1981) High performance liquid chromatographic determination of organic acids in dairy products. J. Food Sci. 46, 52±57. Martens, H. and Nñs, T. (1989) Multivariate Calibration. John Wiley and Sons, Chichester. Muir, D. D. and Hunter, E. A. (1991) Sensory evaluation of Cheddar cheese: order of tasting and carryover e€ects. Food Quality Preference 3, 141±145. Muir, D. D. and Hunter, E. A. (1992) Sensory evaluation of fermented milks: vocabulary development and the relations between sensory properties and composition and between acceptability and sensory properties. J. Soc. Dairy Technol. 45, 73±80. Muir, D. D., Tamime, A. Y. and Hunter, E. A. (1995) Sensory properties of Kishk: comparison of products containing bovine and caprine milk. J. Soc. Dairy Technol. 48, 123±127.

Passmore, R. and Eastwood, M. A. (1986) Davidson and Passmore Human Nutrition and Dietetics. 8th edn, pp. 40± 52, 103±131. Churchill Livingstone, Edinburgh. Paul, A. A., Southgate, D. A. T. and Russell, J. (1980) First Supplement to McCance and Widdowson'sÐThe Composition of Foods. HMSO, London. Pomeranz, Y. (1991) Functional Properties of Food Components. Academic Press Inc., New York. Tamime, A. Y. and O'Connor, T. P. (1995) KishkÐa dried fermented milk/cereal mixture. Int. Dairy J. 5, 109±128. Tamime, A. Y., Muir, D. D., Barclay, M. N. I., Khaskheli, M. and McNulty, D. (1997) Laboratory-made Kishk from wheat, oat and barley: 1. Production and comparison of chemical and nutritional composition of Burghol. Food Res. Int. 30, 5, 311±317. Tamime, A. Y. and McNulty, D. (1998) KishkÐa dried fermented milk/cereal mixture: 4 Microbiological quality. Lait (in press). Tamime, A. Y., Barclay, M. N. I., Amarowicz, A. M. and McNulty, D. (1998a) KishkÐa dried fermented milk/cereal mixture: 1. Chemical composition. Lait (in press). Tamime, A. Y., Barclay, M. N. I., McNulty, D. and O'Connor, T. P. (1998b) KishkÐa dried fermented milk/ cereal mixture: 3. Nutritional composition. Lait (in press). Williams, S. A. R., Hunter, E. A., Parker, T. G., Shankland, C. E., Brennan, R. M. and Muir, D. D. (1996) DDASP: a statistical based system for design, data capture and analysis with the sensory pro®ling protocol. In Proceedings 3eÁme Congres Sensometrics, 19±21 June 1996, pp. 48.1±48.3. Nantes.

(Received 19 December 1996; accepted 24 August 1997)