Enhancing molecular property prediction by fusing graph and sequence encoder representations

Enhancing molecular property prediction by fusing graph and sequence encoder representations

연구 분야: Databases

학회: Knowledge and Information Systems

Drug discovery through wet lab experimentation is a time-consuming and expensive process. The high cost of drug development is largely due to research and clinical testing of potential drug candidates that may not achieve regulatory approval. Virtual screening methods offer a computational alternative, but existing methods often rely solely on molecular sequences or graph structures, limiting predictive accuracy. This work introduces a novel fusion learning framework that integrates both molecular graph structure information and molecular sequence contextual information of molecules to enhance molecular representation learning. This is done by combining the graph and sequence embeddings to generate a fused representation that improves generalization. Extensive experiments on benchmark datasets demonstrate that our method significantly outperforms graph-based, sequence-based, and prior fusion approaches. The complete implementation of this work is available at https://github.com/vyshakhgnair/TraGT.