Developing an intelligent data-driven framework for organic photovoltaic research

The current trial-and-error research paradigm is inherently inappropriate for organic photovoltaic (OPV) development because of the diversity in the properties of photovoltaic materials and sophisticated device fabrication conditions. A data-driven paradigm is therefore particularly well-suited to address these challenges. The data-driven research paradigm requires big-data extraction from the literature and high-throughput quantum chemical calculation according to chemical structures and quantitative molecular structure–property relationship (QMSPR). Accordingly, we develop an intelligent data-driven framework (IDDF), leveraging large language models (LLMs), a high-throughput quantum chemistry calculation platform (HQCCP), and explainable machine learning (ML). With the aid of IDDF, we extract structured data from 615 peer-reviewed articles and compute 50 molecular descriptors for 125 Y-series acceptors, forming a QMSPR database linking molecular features to device performance. Using the eXtreme Gradient Boosting (XGBoost)-SHapley Additive exPlanations (SHAP) ML model, IDDF maps molecular substructures to key descriptors and device power conversion efficiency (PCE). The results quantitatively demonstrate the influence of each building block of the Y-series acceptors on the final PCE values with explicit quantum chemical explanations. Our framework shifts OPV research from intuition-based design toward a knowledge-guided, predictive mode, providing a foundational step toward autonomous material discovery and enhancing the competitiveness of OPVs among emerging photovoltaic technologies.