Machine Learning Framework for Automated SSR Marker Quality Assessment in Horsegram (Macrotyloma uniflorum)

Background: Simple Sequence Repeat (SSR) markers are crucial tools for molecular breeding and genetic diversity studies in legumes. Traditional SSR marker development relies on subjective quality assessment methods, which are time-consuming, costly and prone to inconsistency. The lack of quantitative frameworks for predicting marker quality limits the efficiency of breeding programs and genomic studies. To develop a comprehensive machine learning framework for automated SSR marker quality prediction, identify key determinants of marker success through quantitative analysis and establish evidence-based design principles for efficient marker development in legume crops. Methods: We engineered 15 predictive features from primer design parameters and SSR structural characteristics of 99 horsegram SSR markers. Three machine learning algorithms (Random Forest, Support Vector Regression and Neural Network) were trained and validated using cross-validation. Feature importance analysis quantified the contribution of each parameter to marker quality prediction. Result: Random Forest achieved optimal performance (R² = 0.378, MAE = 8.106) with 37.8% explained variance in marker quality prediction. Feature importance analysis revealed primer compatibility factors as dominant predictors: GC content balance between primers (35.6% importance) and melting temperature compatibility (32.2% importance). SSR structural features contributed 22.9% importance, with motif complexity (8.6%) and motif length (8.5%) being most significant. Cross-validation confirmed robust model performance (CV R² = 0.342 ± 0.089) across different data subsets.