LoRA-Chem: Modular Machine Learning for Multitask Prediction in Organic Reactions

Ben Gao, Penghui Li, Di Zhang, Qian Tan, wanhao liu, Xunzhi Wang, Junxian Li, Shufei Zhang, Dongzhan Zhou, Yuqiang Li, Guoyin Yin

Abstract: The prediction of organic chemical reactions has historically presented significant challenges owing to the inherent complexity and mechanistic diversity of reaction processes. In this work, we present LoRA-Chem, an innovative modular framework that demonstrates remarkable performance in predicting individual organic reactions while achieving unprecedented multitask capacity. This unified architecture enables accurate forecasting of diverse reaction types through a single integrated model - a critical advancement that addresses longstanding limitations in traditional prediction methodologies. Importantly, our investigation reveals that strategic improvements in the foundational model architecture can substantially enhance the system's predictive accuracy. The demonstrated versatility and scalability of the LoRA-Chem framework suggest transformative potential for next-generation reaction prediction systems, offering a robust platform for developing sophisticated artificial intelligence solutions in synthetic chemistry.

External IDs:doi:10.26434/chemrxiv-2025-p7sxn-v2