TSMGen: Target-Specific Molecule Generation via Higher-Order Structural Dependencies and Context-Aware Bidirectional Fusion

Efficiently designing high-quality molecules targeting disease-relevant targets is a critical challenge. Most existing methods can capture pairwise amino acid relations, neglecting the higher-order relations among multiple amino acids. This paper proposes a target-specific molecule generation framework, namely TSMGen, to comprehensively capture the local and global structural information of the protein pocket by modeling higher-order spatial dependencies both at the atomic and the amino acid levels. Furthermore, we design a context-aware bidirectional fusion module to learn the more detailed structural information about the protein pocket. This module simultaneously attends to features from both the protein pocket and the molecule, fully leveraging the structural information from both to optimize the generation process of targeted molecules, thereby enhancing the quality of generated molecules. Experiments show that TSMGen outperforms state-of-the-art methods in terms of Vina Score, High Affinity, QED, SA and Diversity, and a case study on $\beta$-secretase enzyme further confirms its ability to generate molecules with stronger binding affinity.

Modern drug discovery is expensive and time-consuming, especially when scientists need to design molecules that can precisely bind to disease-related proteins. Artificial intelligence has recently shown promise in helping generate new drug candidates, but many existing methods still struggle to fully understand the complex structure of protein binding sites. In particular, they often focus only on simple pairwise relationships between amino acids and overlook more complicated structural interactions that are important for molecular recognition. In this work, we introduce TSMGen, an AI framework for target-specific molecule generation. Our method studies protein binding pockets at multiple levels, capturing both detailed local atomic information and broader structural patterns across groups of amino acids. It also allows the protein and the generated molecule to continuously exchange information during the design process, helping the model produce molecules that better match the target protein environment. We evaluated our approach on public drug discovery datasets and compared it with several state-of-the-art molecular generation methods. The results show that TSMGen generates molecules with stronger predicted binding ability, better drug-like properties, and higher chemical quality. In a case study involving a protein related to Alzheimer’s disease, our method produced candidate molecules with stronger predicted binding affinity than known reference compounds. These results suggest that TSMGen could help accelerate the early stages of drug discovery by improving the design of candidate molecules for specific disease targets.