Urdbean (Vigna mungo L. Hepper) cultivar characterization based on multiple seed and flour properties and their multi-variate analysis using artificial neural network

Abstract

Urdbean is one of the most highly consumed pulse crops in South Asia. For the last 35 years, several varieties were developed. However, the physical parameters of seeds, functional properties of flour samples and swelling power or solubility of starch samples of urdbean were never studied in varieties systematically released over decades. Hence, the present study was undertaken based on mentioned parameters to formulate artificial neural network (ANN) model for varietal identification. Significant variability was observed for all recorded parameters among the tested urdbean varieties. The mean length, width, and thickness of urdbean seeds was 4.61 mm, 3.26 mm and 3.24 mm and varied one variety to another from 3.94 mm (VBN-6) to 5.20 mm (Vamban 7), 2.70 mm (WBU 109) to 3.63 mm (IPU18-02) and 2.66 mm (WBU 109) to 3.63 mm (IPU18-02), respectively. Similarly, the mean seed volume, mean area of surface, mean area of transverse surface, color properties and sphericity were significantly varied among tested urdbeans. The mean value of water absorption capacity of urdbean starch samples was 4.97 and the mean value of oil absorption capacity of urdbean samples was 1.71. The mean solubility of urdbean starch was 20.82 per cent and swelling power of starch (g/g) of urdbean genotypes had a mean value of 9.06. The mean seed hardness of urdbean genotypes was 5286.80 (N/mm²). The mean values of pick viscosity, other allied parameters were found to be variable among urdbeans tested. Most of the parameters approach R² of 0.90, suggesting an excellent fit of the ANN model.

Graphical Abstract

Introduction

Urdbean or blackgram or uradbean (Vigna mungo L. Hepper) is a widely consumed food legume crop in South Asia. India is the largest grower, it produces about 2.23 million tonnes of urdbean yearly utilizing about 4.11 million hectares of cropped area, however, it has a low productivity of about 538 kg/hectare (Project Coordinator’s Report, 2022). Urdbean production contributes > 10 percent to the total pulses production in India (Project Coordinator’s Report, 2022). Most of the recent urdbean varieties developed were the product of hybridization followed by pedigree selection. However, varieties developed during 1940-50 s were mass selections from local land races or cultivars. T 9 variety was developed by selection from a local land race of urdbean (Sen Gupta et al., 2020). Most of the varieties developed in urdbean are the product of intra- as well as inter-specific hybridization for multiple trait introgression in a single cultivar. Inter-specific hybridization in urdbean breeding has become more common with the application of hormones or other chemicals. During last decade, at least nine urdbean varieties were released which were the products of inter-specific hybridization in India. It is interesting to note that most of the products of inter-specific hybridization were products of mungbean or V. mungo var. silvestris crosses with urdbean (Sen Gupta et al., 2020). It is now evident to utilize other Vigna species for alien gene introgressions into urdbean. Hence, development of multiparental breeding population is a priority area in this crop species. Besides this, a few urdbean varieties were developed using mutation breeding techniques supported by Bhabha Atomic Research Centre, India (Venugopalan & Suprasanna, 2022).

Primary focus of current urdbean breeding programmes is to develop mechanical harvesting suitable or sympodial (soybean type) type varieties in India and IPU 13–1 has been developed by ICAR-IIPR, Kanpur as a main-stem pod bearing urdbean variety. Such plant types help to accommodate a greater number of plants per unit area leading to increased yield. Moreover, this also facilitates mechanical harvesting. Thus, over last 35 years more than hundred urdbean varieties were developed in India. However, these varieties were not characterized based on profiling of different seed bio-physical and functional properties of flour samples, limited number of studies were conducted to characterize urdbean genotypes on different seed quality traits like iron and zinc concentration, anti-nutrient like phytic acid concentration (Sen Gupta et al., 2020, 2022) and for seed bio-physical traits (Sen Gupta et al., 2023). Further, development of artificial neural network (ANN) model for identification of urdbean varieties based on above parameters would provide an opportunity to develop improved varieties in future by targeting specific parameters. This has been successfully used in wheat where an ANN model was developed for classifying varieties based on 11 physical properties of wheat grains (Taner et al., 2018) and for estimating extensograph properties of dough (Abbasi et al., 2012). Hence, the objectives of these studies were to find out the (i) seed bio-physical (i.e. seed length, breadth, height, volume, diameter, 1000 seed weight, bulk density, hardness, color analysis, and other allied parameters), functional (water absorption capacity, oil absorption capacity), pasting properties and swelling power and solubility of urdbean flour samples, (ii) spectrophotometer (FTIR) based profiling of urdbean samples, and (iii) to develop a artificial neural network model for predicting physico-chemical properties of urdbean cultivars, with a focus on specific traits relevant to urdbean breeding programs.

Materials and methods

Plant materials and field trial

A field trial comprising of 16 popular varieties released over 35 years, and one advanced breeding line of urdbean (Table 1), was conducted at main farm of ICAR-IIPR, Kanpur in RCBD with 3 replications following recommended agronomic practices. Seeds were harvested, cleaned, dried, and stored at 4 °C for further analysis.

Table 1 Details of seventeen urdbean genotypes used in the study

Analysis of physical properties

Ten seeds were randomly chosen to measure their length (L), breadth (B), and height (H) with a Vernier caliper having a precision of 0.01 mm. Various parameters related to the size and shape of the seeds, such as volume (V, Eq. 1), geometric diameter (D_g, Eq. 2), arithmetic diameter (D_a, Eq. 3) sphericity (S, Eq. 4), area of surface (A_f,, Eq. 5) and area of transverse surface (A_t, Eq. 6) were calculated based on the length, width, and thickness. The bulk density (BD, Eq. 7), angle of repose (Eq. 8) and thousand grain weight were recorded as described by Sharanagat & Goswami (2014).

$$\text{V}=\left(\frac{\uppi }{6}\right)*\text{L}*\text{B}*\text{H}$$

(1)

$${\text{D}}_{\text{g}}={\left(\text{L}*\text{B}*\text{H}\right)}^\frac{1}{3}$$

(2)

$${\text{D}}_{\text{a}}=\left(\frac{\text{L}+\text{B}+\text{H}}{3}\right)$$

(3)

$$\text{S}={(\left(\text{L}*\text{B}*\text{H}\right)}^\frac{1}{3}/\text{L})*100$$

(4)

$${\text{A}}_{\text{f}}=\left(\frac{\uppi }{4}\right)*\text{L}*\text{B}$$

(5)

$${\text{A}}_{\text{t}}=\left(\frac{\uppi }{4}\right)*\text{H}*\text{B}$$

(6)

$$\text{BD}=\frac{\text{weight of seeds }\left(\text{kg}\right)}{\text{volume of container }({\text{m}}^{3})}$$

(7)

$$\text{Angle of repose}={\text{tan}}^{-1}\left(\frac{2\text{h}}{\text{d}}\right)$$

(8)

h = Height of the pile

d = Diameter of the pile of seed

Color analysis

The color analysis of various urdbean varieties was performed using a Chroma Meter (Konica-Minolta, CR-400, Japan) as described by Shivani et al. (2023). The color parameters, including L (lightness/darkness), a (redness/greenness), and b (yellowness/blueness) were measured and total color difference (∆E, Eq. 9), browning index (BI, Eq. 10), chroma (Eq. 11) and Hue (Eq. 12) values were calculated using the respective equations given below.

$$\Delta \text{E}= \sqrt{{({\text{L}}_{1}-{\text{L}}_{0})}^{2}+{({\text{a}}_{1}-{\text{a}}_{0})}^{2}+{({\text{b}}_{1}-{\text{b}}_{0})}^{2}}$$

(9)

$$BI=\frac{X-0.31}{0.17}\times{100}$$

(10)

$$Chroma=\sqrt{{a}^{2}+{b}^{2}}$$

(11)

$$Hue={\text{tan}}^{-1}\frac{b}{a}$$

(12)

$$\text{x }= \frac{({\text{a}}_{1}+1.75{\text{L}}_{1})}{5.46 {\text{L}}_{1}+{\text{a}}_{1}-3.01{\text{b}}_{1}}$$

(13)

where,

${\text{L}}_{1}, {\text{a}}_{1}, {\text{b}}_{1}$ are the values for Urdbean sample

${\text{L}}_{0}, {\text{a}}_{0}, {\text{b}}_{0}$ are the values of white slab

Hardness of seed

The hardness of seventeen urdbean genotypes including 16 varieties was analyzed by a texture analyzer (TA-XT, Stable Micro Systems, United Kingdom). Each sample was carefully positioned on the platform of texture analyzer to ensure accurate measurements, and the compression mode was employed to measure the hardness. The test speed (1 mm/sec) was maintained during compression. Following the completion of the compression, the post-test speed was set to 10 mm/sec. The target mode for the analysis was strain (50%) and a trigger force of 10 g was set as the threshold force required to initiate the compression test (Sharanagat et al., 2018).

Functional properties

Water absorption capacity (WAC) and Oil absorption capacity (OAC)

A 3 g sample of urdbean flour was accurately weighed and transferred into a pre-weighed centrifuge tube followed by the addition of distilled water (25 mL). The mixture of sample and water was vigorously shaken for 5 min to ensure thorough dispersion and contact between the starch particles and water. After shaking, the mixture was allowed to stand undisturbed for 30 min at a controlled temperature of 25 °C. This allowed the sample to absorb the water. Then the sample was centrifuged at 3000 g for 25 min and the supernatant was carefully drained from the centrifuge tube. The centrifuge tube with the sediment was then dried in a hot air oven at 50 °C for 25 min. The WAC of urdbean sample was determined using Eq. (14) (Kaur et al., 2024).

$$WAC\left(\frac{g}{g}\right)=\frac{Weight\,of\,absorbed\,water}{Initial\,weight\,of\,sample}$$

(14)

The sample (0.5 g) and soybean oil was mixed in a pre-weighed centrifuge tube. The sample-oil mixture was allowed to incubate for 30 min (25 °C) followed by centrifugation at 3000 g for 25 min. After centrifugation supernatant was drained and the centrifuge tube was inverted over tissue paper for 25 min to drain the unabsorbed oil. The OAC was calculated using Eq. (15) (Kaur et al., 2024).

$$OAC(\frac{g}{g})=\frac{Weight\,of\,absorbed\,oil}{Initial\,weight\,of\,sample}$$

(15)

Swelling power and solubility

Swelling power (SP) and solubility (S) of urdbean flour samples were measured using the method described by Hong et al. (2021) with slight modifications. A sample mixture was prepared by combining 1 g (dry basis) of sample with 30 mL of distilled water. The sample mixture was heated in a water bath at 95 °C with continuous stirring for 30 min followed by cooling at room temperature. The cooled sample mixture was then subjected to centrifugation (3500 rpm, 20 min) and the supernatant obtained from centrifugation was carefully collected, and dried (105 °C for 24 h). The swelling power (SP) and solubility (S) was calculated using Eq. (16) and (17), respectively.

$$Swelling\,Power\,\left(\frac{g}{g}\right)=\frac{Weight\,of\,sample\,paste}{Initial\,weight\,of\,sample}$$

(16)

$$Solubility\,\left(\frac{g}{g}\right)=\frac{Weight\,of\,supernantant\,(Dry\,mass)}{Initial\,weight\,of\,sample}$$

(17)

The above-mentioned procedures were repeated for all the varieties.

Pasting properties

The pasting parameters of 17 different genotypes of urdbean flour samples were determined using an Anton Paar MCR 52 Rheometer (Austria). A sample suspension was prepared by adding 2 g of urdbean flour in 16 mL of distilled water. Manual agitation was performed to ensure the even distribution of sample within the suspension. The suspension was subjected to heating, starting from 50 °C and increasing to 95 °C at a rate of 6 °C per minute. Once the temperature reached 95 °C, it was held constant for 5 min to ensure complete gelatinization of the starch. After the holding period at 95 °C, the suspension was cooled from 95 °C to 50 °C at a rate of 6 °C per minute. Once the temperature reached 50 °C, it was maintained for 2 min (Chakraborty et al., 2024).

FTIR spectrophotometer analysis

The Fourier transform infrared (FT-IR) spectrogram of 17 different genotypes of urdbean was examined using an ALPHA Bruker FTIR spectrometer. To conduct the analysis, powdered samples were carefully positioned on an attenuated total reflection plate (ATR), and FTIR spectra were recorded within the range of 4000–400 cm^-1 (24 scans). Before each measurement, the ATR plate was manually cleaned using isopropyl alcohol and a background reading was taken to eliminate any potential errors or interferences (Srikanth et al., 2021). The principal component analysis (PCA, Originpro 2018, USA) was performed on the major spectral bands of different samples to further distinguish the varieties-based variations in absorbance (Lamas et al., 2021).

Statistical analysis and artificial neural network

All the analysis including calculation of mean, and mean separations were performed using Originpro 2018 (Origin Lab Northampton, Massachusetts, USA). One-way anova turkey test (p < 0.05) was performed using the statistical analysis software SPSS25.

The Artificial Neural Network comprises an input layer, a hidden layer, and an output layer, as provided in (Fig. 1B). The selection of input and target variables was done based on correlation analysis (Fig. 1A). In the context of predicting a particular property, those data are taken as input whose correlation coefficient exceeds + 0.5 or falls below −0.5 with that particular property. The ANN analysis was done using R2020b (MATLAB 9.9 Natick, Massachusetts, USA). As per the amount of data in input and target, the whole data set was divided into training (70%), validation (15%), and testing (15%). The number of hidden layers is 10 for each prediction. The training of the data set was done using the Levenberg–Marquardt algorithm, and the data was trained until MSE (mean squared error), and the regression R-value was not close to 1. The number of epochs, representing the number of times the entire training dataset is passed through the model, was determined automatically. For the validation R², Χ² and RMSE were calculated by the experimental value and predicted value using Origin 2018 (Origin Lab Northampton, Massachusetts, USA).

Fig. 1

A Correlation analysis of different pysico-functional and pasting properties of urdbean genotypes B) The proposed ANN topology

Results and discussion

Physical properties

The knowledge of physical properties of grains is important for the planning of different processing operations, packaging and transportation, and design and development of processing equipment (Babić et al., 2011). In the present study, different physical properties of urdbean seeds were analyzed and shown in Table 2. The mean length, width, and thickness of urdbean seeds was 4.61 mm, 3.26 mm and 3.24 mm and varied one variety to another from 3.94 mm (VBN-6) to 5.20 mm (Vamban 7), 2.70 mm (WBU 109) to 3.63 mm (IPU18-02) and 2.66 mm (WBU 109) to 3.63 mm (IPU18-02), respectively. In an earlier study, length, breadth, and thickness of seeds were in the range of 4.66–5.11 mm, 3.71–3.79 mm and 3.20–3.29 mm, respectively between two urdbean cultivars (Wani et al., 2013). Thus, a wider range of variation for these parameters in the present study could be due to the use of 16 diverse varieties of urdbean. The variation in dimension led to the change in geometric mean diameter and arithmetic mean dimension and varied in the range of 3.27 mm (WBU 109) to 4.00 mm (IPU18-02) and 3.35 mm (VBN-6) to 4.04 mm (IPU18-02), respectively. Similarly, the mean seed volume, mean area of surface, mean area of transverse surface and sphericity was 25.68 mm³, 11.83 mm², 8.33 mm², and 79.27%, and varied in the range of 18.25 mm³ (WBU 109) to 33.59 mm³ (IPU18-02), 9.53 mm³ (VBN-6) to 13.93 mm³ (Pant Urd 31), 5.64 mm³ (WBU 109) to 10.37 mm³ (IPU18-02) and 67.32% (WBU 109) to 84.67% (IPU10-26), respectively. The variation in physical properties is observed directly correlated with the genetic constitution of the genotypes. Therefore, it can be positively used for the identification and development of handling and grading equipment for the urdbean. Surface properties, volumetric density and 1000 seed weight are also important characteristics for the transportation and storage of grain. The mean value of angle of repose, bulk density and 1000 grain weight was 0.42, 0.81 kg/mm³ and 31.86 g and varied in the range of 0.32 (IPU10-26) to 0.46 (Mash 114), 0.76 kg/mm³ (IPU11-02) to 0.87 kg/mm³ (IPU10-26) and 27.34 g (Shekhar 2) to 39.02 g (Vamban 7), respectively. Similar variation was also recorded for different seed traits among individuals of two mapping populations involving intra and inter-specific crosses (Sen Gupta et al., 2023). In the first mapping population, 100 seed weight was ranged from 2 to 5 g with a mean value of 4 g. Seed length ranged from 4 to 5 mm with a mean value of 5 mm. The mean seed width was 4 mm. for seed width. Seed thickness ranged from 3 to 4 with a mean value of 3 mm. Geometric mean of diameter was varied between 3—4 mm with a mean value of 4 mm. Sphericity of seed had a mean value of 16.2 mm³. Seed surface area ranged from 129 to 230 mm² and mean value was 181 mm². In the second mapping population (IPU11–02 × Pant M 6) also similar variations were observed. (Sen Gupta et al., 2023). The variation in the physical properties of millet grains with change in varieties has also been reported by Adebowale et al. (2012).

Table 2 Physical properties of grains of tested seventeen urdbean genotypes

Color analysis

The color of the urdbean varied with the genotypes (Table 3). Mean values of ‘L’, ‘a’ and ‘b’ was 24.16, 1.05 and 3.56, and varied in the range of 22.03 (IPU94-1) to 30.03 (Shekhar 2), 0.26 (IPU10-26) to 2.54 (WBU-108) and 1.50 (Pragati) to 7.11 (WBU-108), respectively. The mean value of Chroma, Hue, total color difference (ΔE) and BI was 3.74, 73.87, 67.31 and 18.91, and varied in the range of 1.70 (Pragati) to 7.56 (WBU-108), 58.28 (IPU11-02) to 84.36 (WBU-109), 61.46 (Shekhar 2) to 69.05 (KUG-479) and 8.84 (IPU10-26) to 38.08 (WBU-108) respectively. Color is one of the deciding parameters for consumer acceptance. A low value for chroma and a high value for lightness are desired to meet the consumer preference (Kavitha et al., 2013). Usually, a lower value (0–50) of ‘L’ value indicates dark and a higher value (51–100) indicates lighter color. A positive number of ‘a’ value indicates red, and a negative number indicates green. Similarly, a positive ‘b’ value indicates yellow, and a negative value indicates blue. Hue angle determines the darkness or lightness of the shade. It is evident from the data (Table 3) that the positive values of ‘a’ and ‘b’ show that it has a reddish yellow shade and low range of ‘L’ (22–30) indicated dark colored flour. A high ‘L’ value of 75 and the hue angle value approaching 90 degrees represent lighter shade. This will be beneficial to maintain the white shade while mixing with other cereals. Comparatively higher ‘L’ value (30) of urdbean variety Shekhar-2 as this urdbean variety has dull green seed coat color and others have dull or shiny black seed coat color. The color of the pulse flour also affects the quality of end product. The incorporation of pulse flour led to an increase in crust darkness of bread while marginable change has been observed in crumb color (Nkurikiye et al., 2023). The lighter color urdbean flour can be used for the development of nutritive bread to avoid the dark color development. In order to find urdbean flour with better or suitable quality for end use, testing of a large number of urdbean germplasm will be required.

Table 3 Details of flour color and functional properties of tested urdbean genotypes

Functional properties

In the present study, different functional properties of flour samples were analyzed for all tested seventeen urdbean genotypes (Table 3). Mean water absorption capacity (WAC) of urdbean starch samples was 4.97 with a range from 3.17 (IPU11-02) to 6.43 (Vamban 7). Earlier 18 different pulses along with rice and corn sample was studied for WAC and found that among the pulses, adzuki bean had higher water holding capacity (3.74) followed by Light Red Kidney Bean (2.06), while other pulses samples had WAC < 2 (Lentil 1.52, Garbanzo 1.87, Field Peas 1.65, Great Northern Bean 1.32) (Sangokunle et al., 2020). Rice (1.86) and corn (1.34) samples were having very low water holding capacity than pulses. These findings showed that urdbean starch has more WAC than cereals and other discussed pulses. In the present study, high WAC (~ 5) was observed in both old and newer (< 10 years) urdbean varieties, except IPU11-02. Oil absorption capacity (OAC) of urdbean samples ranged from 1.59 (KUG-479) to 1.95 (Azad Urd 1) with a mean value of 1.71. Sangokunle et al. (2020) reported high OAC (2.54) in rice compared to pulses (Adzuki bean 0.78, Field Pea 0.86, Grabanzo 1.35, Great Northen Bean 1.16), and corn (0.79). However, in our study urdbean varieties Azad Urd 1 (1.95) and IPU17-1 (1.90), had higher OAC than corn (0.79) but less than Rice (2.54). However, use of a greater number of genetic resources of urdbean in future study can possibly identify urdbean genotypes with OAC more than rice.

Swelling power (SP) and solubility of starch are controlled by amylose and amylopectin and confined to amorphous and crystalline regions of the starch granule (Kaur et al., 2018; Ratnayake et al., 2002). The mean solubility of urdbean starch was 20.82% with a range from 9.50% (IPU02-43) to 27.32% (Vamban-7) (Table 3). In general, high solubility of starch was observed among the older urdbean varieties (varieties released before year 2010), while low solubility of starch was observed in new varieties that released after year 2010. However, exceptionally low solubility (9.5%) was observed in an old variety IPU02-43, that was released in year 2009 and high solubility of starch was observed in IPU17-1 (26.65%), one of the recently released (2021) urdbean varieties. Kaur et al. (2018) reported starch solubility of different cereals while increasing temperature gradually from 55 °C to 95 °C and at 95 °C the starch of rice, wheat, oats, maize, barley, sorghum, and millets were 8.56, 6.77, 12.3, 9.30, 5.97, 10.26 and 13.56 per cent. In this study it was observed that urdbean starch solubility is manifold higher than cereal starch solubility. Swelling power of starch (g/g) of urdbean genotypes ranged from 7.35 (Shekhar-2) to 11.50 (Azad Urd-1) with a mean value of 9.06. Kaur et al. (2018) reported the starch swelling power (g/g) of rice, wheat, oats, maize, barley, sorghum, and millets at 95 °C were 11.56, 12.30, 13.12, 10.29, 9.31, 12.67 and 14.20, respectively. Hence, from the present study conclusion can be made that few urdbean varieties [Azad Urd 1 (11.50 g/g), WBU-109 (11.15 g/g), IPU10-26 (10.07 g/g)] had the swelling power at par with rice, maize, and barley. Inclusion of more urdbean genotypes may lead to the identification of urdbean starch with at par or higher SP than millets.

Mechanical properties

The mechanical properties of urdbean are listed in Table 2. The mean hardness of urdbean genotypes was 5286.80 (N/mm²) and varied in the range of Vamban-7 (7841N/mm²) to IPU17-1 (1876 N/mm²). Most of the recently released urdbean varieties from IIPR, Kanpur [IPU10-26 (4768N/mm²); IPU13-1 (3434N/mm²), IPU11-02 (2683N/mm²); and IPU17-1(1876N/mm²)] has lower (manifold) seed hardness which is a signal of remarkable progress in urdbean breeding to avoid the ‘hard to cook’ (El-Tabey, 1992) phenomenon and easy grinding. Since the analysis was performed for freshly harvested seeds, the hardness of these varieties needs to be tested under different storage durations. Biochemical profiling of these high and very-low hardness exhibiting genotypes should be performed to find a chemical basis to such high variability in terms of seed hardness. For detailed study, involving micro-structure analysis, phenolics content, phytic acid concentration should be investigated before deriving the conclusion as a possible reason for very low seed hardness as observed among few genotypes (Pirhayati et al., 2011).

Pasting properties

The pasting of urdbean flour genotypes have been analyzed and shown in Table 4. The mean values of pick viscosity, holding viscosity, breakdown viscosity, final viscosity, setback from peak, setback from trough and pasting temperature was 1542.06 cP, 1025.27 cP, 526.44 cP, 1232.16 cP, 349.21 cP, 196.68 cP and 72.29 °C and varied in the range of 1254.33 (IPU11-02) to 1834 cP (IPU10-26), 763.57 (Vamban-6) to 1254.33 cP (Vamban-7), 291.70 (WBU-109) to 825.30 cP (IPU10-26), 791.15 (Vamban-6) to 1576 cP (IPU94-1), 6.68 (Shekhar-2) to 746.50 cP (IPU10-26), 10.39 (MASH-114) to 563 cP (IPU94-1) and 62.31 °C (Vamban-6) to 76.53 °C (IPU94-1), respectively. The different old and newly developed varieties had variable peak viscosity. The varieties like Shekhar-2 and IPU11-02 had low peak viscosity and higher peak viscosity was observed for varieties like IPU02-43 and IPU10-26. Peak viscosity was found to be linked with water binding capacity of starch which occurs at equilibrium point between swelling. This causes an increase in viscosity (Ocheme et al., 2018; Sanni et al., 2001). The trough, breakdown, final, and setback viscosities also corresponded to the peak viscosity data of seventeen urdbean starch samples. All recently released urdbean varieties (less than 10 years) had high pasting temperatures (> 70 °C) except the Vamban-6 (62 °C). Among the varieties from IIPR, Kanpur most of the varieties had pasting temperatures higher than 75 °C except the IPU18-02 (69.69 °C) and IPU17-1 (71.6 °C). Starch of different pulses have different properties (Singh et al., 2017). Pulse starches have low swelling and are difficult to shear compared to cereals (Singh, 2010; Singh et al., 2008). However, from the present experiment few urdbean varieties [IPU10-26 (1834 cP), IPU02-43 (1800 cP), WBU 108 (1760 cP)] were found to have better pasting properties than others. Blending pulses with cereal flour surely may cause the reduction of glycemic index of the product. Hence, pulse starch is used for diabetic consumers for its use in different food products (Singh et al., 2017) and urdbeans can surely contribute to this.

Table 4 Pasting properties of urdbean samples of seventeen different genotypes

FTIR analysis

The FTIR spectra of seventeen genotypes of urdbean exhibited a similar absorbance pattern (Fig. 2). The overall spectral range for all varieties was analyzed and the clusters were categorized based on the absorbance of major bands. The strong band at 1010–1020 cm⁻¹ is due to C-O stretching vibrations, which can be present in various functional groups such as ethers, esters, or alcohols (Taheri et al., 2020). A weak band at 1153 cm⁻¹ could be associated with C-N stretching vibrations, commonly found in amines or amides. The band at 1239 cm⁻¹ is due to the C-N stretching vibrations of proteins. The band at 1530 cm⁻¹ is attributed to the N–H bending vibrations of amides II, which are functional groups containing both a carbonyl group (C = O) and an amino group (NH2) (Carbonaro et al., 2008). The peak at 1648 cm⁻¹ is a characteristic band for C = O stretching vibrations, indicating the presence of carbonyl groups (Ahmed et al., 2009). The broad band near 3300 cm⁻¹ is generally associated with O–H stretching vibrations due to the presence of hydroxyl groups in alcohols or phenols. The FTIR spectra of various urd genotypes reveal noticeable differences in absorption band intensity between the genotypes. According to Beer-Lambert's law, the intensity of absorbance is directly proportional to the concentration of the absorbing species (Kumar & Gill, 2018). This implies that a higher intensity corresponds to a greater concentration of the functional group in the sample. For instance, the genotypes IPU94-1, IPU18-02, and Vamban-6 exhibit lower intensity of absorbance peak as a result a lower concentration of C-O groups, with absorption bands in the 1010–1020 cm⁻¹ range was observed. The intensity of the O–H group is higher in IPU02-43 and Vamban-6 compared to other genotypes, indicating a greater abundance of O–H groups in these samples. The intensity of an absorption band is also influenced by bond polarity; bonds with higher polarity generate more intense absorption bands (Barth, 2000). Therefore, the observed differences in intensity among the genotypes may be attributed not only to variations in concentration but also to changes in the polarity of the bonds.

Fig. 2

FTIR analysis of the seventeen urdbean genotypes flour samples

Figure 3 shows a principal component analysis (PCA) biplot, having verities as Scores and band wavenumber as Loading plot provides further insight. It represents a total of 89.79% (PC1-66.87% and PC2- 22.92%) of total variations. The plot shows three distinct clusters formed by the 17 varieties of urdbeans. Green group contains varieties like Vamban-6, IPU18-02, and IPU94-1 having a minor band at 2347 cm⁻¹ indicating residual CO₂ from the atmosphere or from metabolic processes. However, this band was absent in most of the samples. The second group (purple circle) includes WBU-109, IPU02-43, KUG-479 and IPU13-1. The scores of this group have bands at 1540 (N–H bending in amide II band or C = C stretching in aromatic rings), 1648 (C = O stretching), and 1402 cm⁻¹ (C-H bending in alkanes or methyl groups). These bands can be linked to higher protein contents and to the aromatic compounds (Aldehydes, Ketones, etc.). The third group (orange circle) contains varieties like Pragati, Shekhar-2, Vamban-7, and WBU-108 with band at 1011, 1153, 1237 and 3283 cm⁻¹ (O–H stretching). The bands at 1011 and 1153 cm⁻¹ strongly represent C-O stretching in alcohols, ethers, and esters or C-N stretching in amines.

Fig. 3

Principal componenet analysis of FTIR analysis of seventeen urdbean genotypes flour samples

Artificial neural network based analysis

The supplementary Table S1 represents the experimental values, predicted values, and their statistical verification based on R² (coefficient of determination), Χ² (Chi-square), and RMSE (Root mean square error). The high values of R² across most parameters indicate a strong correlation between the experimental and predicted data. Particularly parameters like V, D g, D a, A f, A t, L, a, b, C, dE, BI, PV, BV, SFP, and SFT, where R² approaches 0.90, suggesting an perfect fit of the ANN model to the recorded data. The parameters such as Geometric Diameter and Arithmetic Diameter exhibit extremely low chi-square values, asserting the precision of the ANN model in capturing the variability in these characteristics. For most parameters, the RMSE values are relatively low, emphasizing the overall accuracy of the ANN model. However, for the percent of sphericity (Φ), Bulk Density (BD), Final viscosity (FV), and Water Absorption Capacity (WAC), the R² values are comparatively lower, indicating that the model may have limitations in accurately predicting these parameters. The parameter Percent of sphericity (Φ) exhibits a correlation (R² = 0.81) between the predicted and experimental values, indicating that the ANN model captures a substantial portion of the variability in sphericity. However, the chi-square value of 3.227 suggests a moderate level of discrepancy between the observed and predicted values. The relatively higher RMSE of 1.796 indicates that there might be larger errors in predicting sphericity compared to other parameters. The ANN model for Bulk Density (BD) with an R² of 0.7 indicates a good but not perfect fit to the experimental data. The very low chi-square value of 0.00013 suggests a strong agreement between the model predicted values and experimental values. The low RMSE of 0.011 further supports the model’s ability to predict bulk density accurately. Despite the slightly lower R², the model’s precision in predicting bulk density makes it a valuable tool for applications requiring this parameter. The ANN model for Water Absorption Capacity demonstrates a moderate correlation (R² = 0.73) between predicted and experimental values. The chi-square value of 0.10752 suggests a reasonable fit, while the RMSE of 0.328 indicates a moderate average magnitude of errors. The ANN model for Final Viscosity exhibits a good overall fit with an R² of 0.78, but the relatively high chi-square value of 9628.343 and RMSE of 98.124 indicate significant deviations in predicting this parameter. Another study showed that the response curves generated by the ANN models were more useful than simple correlation coefficients or coefficients in multiple regression equations (Miao et al., 2006). Similarly, ANN was reported to be the best performing tool for the prediction of rice quality including grain physical parameters (Sampaio et al., 2021). Similarly, in wheat ANN was used for predicting extensograph generated data of dough quality (Abbasi et al., 2012). Thus, the present study would be a landmark study for developing machine learning model for physico-chemical properties of urdbean cultivars specific traits and their use in urdbean breeding program.

Conclusion

A wider range of variation was observed for physical parameters of urdbean genotypes in the present study could be due to the genetic diversity of urdbean varieties. Color analysis facilitated the easy identification of light-colored seeds of urdbean genotypes, this will be more useful for determining the suitability urdbean flour samples in baking applications, where mostly lighter colored flour have less impact on the final color development of the baked product. Here, it also showed that urdbean starch has manifold higher WAC than cereals and other pulses. Higher WAC was recorded in the case of both old and new varieties. Tested urdbean genotypes had OAC less than what was reported in rice, however, inclusion of greater number of genetic resources of urdbean in future study can possibly identify urdbean genotypes with OAC more than rice. In general, high solubility of starch was observed among the older urdbean varieties (varieties released before year 2010), compared to new varieties with few exceptions. Urdbean varieties [Azad Urd 1 (11.50), WBU-109 (11.15), IPU10-26 (10.07)] had the swelling power at par with rice, maize, and barley but lower than millets. Inclusion of more urdbean genotypes may lead to the identification of urdbean starch with at par or higher SP than millets. In addition, most of the recently released urdbean varieties from IIPR, Kanpur [IPU10-26 (4768 N/mm²); IPU13-1 (3434 N/mm²), IPU11-02 (2683 N/mm²); IPU17-1(1876 N/mm²)] has lower (manifold) seed hardness which will improve the cooking quality. However, the analysis was performed for freshly harvested seeds, the hardness of these varieties needs to be tested under different storage durations. The different old and newly developed varieties had variable peak viscosity. In terms of peak viscosity, a few urdbean varieties [IPU10-26 (1834 cP), IPU02-43 (1800 cP), WBU 108 (1760 cP)] were found to have better pasting properties than others. All recently released urdbean varieties (less than 10 years old) had high pasting temperatures (> 70 °C) except the Vamban-6 (62 °C). Further, FTIR spectrum based PCA analysis clustered urdbean genotypes into different clusters. Consequently, ANN model was developed using recorded physico-chemical parameters of the urdbean genotypes and the high values of R² across most parameters indicated a strong correlation between the experimental and predicted data. In case of most of the parameters, R² approaches 0.90 along with low chi-square value and error, suggesting an excellent fit of the ANN model to the data. This model can further be used in the urdbean varietal identification utilizing data of physico-chemical parameters of tested samples. The identified contrasting genotypes for the discussed traits can be further used in hybridization programmes to develop mapping/breeding populations for the identification of genetic control of the traits as well as for marker-assisted selection in the breeding programmes. Present study also widens our scope to conduct a multiple environment experiment in the future to study the environmental effects on these traits.