Modeling Chickpea Productivity with Artificial Image Objects and Convolutional Neural Network.

Area of Science:

• Agricultural Science
• Genetics
• Computational Biology

Background:

• Chickpea is a vital global crop with increasing dietary importance.
• Predicting chickpea productivity traits is crucial for agricultural advancement.
• Genomic data offers potential for enhancing crop breeding strategies.

Purpose of the Study:

• To develop a predictive model for two key chickpea productivity traits: thousand-seed weight (TSW) and number of seeds per plant (SNpP).
• To identify important genomic regions and genes influencing these traits using advanced computational methods.
• To demonstrate the utility of genomic prediction in chickpea breeding programs.

Main Methods:

• Genomic data was encoded into Artificial Image Objects.
• A Convolutional Neural Network (CNN) and dictionary learning with sparse coding were used for feature extraction.
• Extreme Gradient Boosting (XGBoost) was employed for regression-based prediction of TSW and SNpP.
• Dense regression attention maps were utilized to identify important genomic factors.

Main Results:

• The developed model achieved prediction accuracies of 84-85% for both TSW and SNpP.
• Significant genomic regions associated with SNpP were identified within 34 genes.
• Significant genomic regions associated with TSW were identified within 49 genes.
• The attention maps highlighted key SNPs contributing to trait prediction.

Conclusions:

• Genomic prediction models, like the one developed, can accurately predict chickpea productivity traits.
• This approach facilitates the identification of genes influencing important agronomic traits.
• The model supports breeding programs in leveraging genetic diversity for developing improved chickpea varieties with desired phenotypes.