Machine Learning Framework for Automated Simple Sequence Repeat (SSR) Marker Quality Assessment in Pigeonpea (Cajanus cajan L.)

Simple sequence repeat (SSR) markers are essential for molecular breeding in pigeonpea (Cajanus cajan L.) compared to traditional development methods, which exhibit 30–70% marker failure rates. Moreover, no quantitative frameworks exist for predicting marker quality in this orphan legume crop. Therefore, the objective of this study was to develop the first machine learning framework for automated SSR marker quality prediction in pigeonpea and identify key determinants of marker success.

Fifteen predictive features were engineered features from 2,770 computationally designed pigeonpea SSR markers, including primer design parameters, compatibility metrics and SSR structural characteristics. Quality scores (0-100) integrated predicted amplification success, specificity and polymorphism likelihood. Three algorithms (Random forest, support vector regression, neural network) were trained using five-fold cross-validation.

Support vector regression (SVR) achieved the best performance (R² = 0.487, MAE = 2.341, CV R² = 0.441±0.032). Feature importance analysis revealed primer compatibility as the dominant predictor (72.9% combined importance): melting temperature difference (25.2%), average melting temperature (23.0%) and primer length difference (17.8%). The framework enables prioritization of top 554 markers (20%), potentially reducing experimental validation costs by 40–60%. This study presents the first quantitative framework for SSR marker quality assessment in pigeonpea. The finding that primer pair compatibility accounts for 72.9% of predictive importance fundamentally shifts understanding from individual primer optimization to primer pair harmony. This validated framework provides scalable methodology for resource-limited breeding programs and is transferable to other orphan legume species.