Machine learning-driven discovery of high-performance hole-conducting organic materials for solar cells and synthetic accessibility assessment

This study presents a comprehensive framework that integrates ML algorithms with synthetic accessibility assessments to facilitate the identification of novel materials with superior charge transport properties. Over 40 learning models were evaluated and tested to accurately predict the hole mobility, random forest regressor was identified as the most effective model (r-squared value of 0.53). 20 thousand organic compounds are designed. Their synthetic accessibility is measured and about 3 thousand compounds that are difficult to synthesize are removed. Hole mobility of remaining compounds is predicted using best ML model. The generated compound space was visualized through dimension reduction method. 30 compounds with highest hole mobility are selected.