bd mol

A deep learning framework for generating drug-like molecules using bidirectional SMILES modeling.

BDMol leverages a bidirectional Transformer architecture to optimize molecular design through chemical language modeling. By processing SMILES strings from both directions, the system captures complex structural dependencies that unidirectional models miss. It excels in scaffold hopping and lead optimization, producing valid chemical structures with high drug-likeness scores (QED > 0.80) and targeted synthetic accessibility. The framework is built on PyTorch and integrates seamlessly with RDKit for rapid property prediction and library generation.

https://github.com/XinhaoLi74/BDMol