Drug-target affinity (DTA) prediction is crucial for drug discovery. While protein evolutionary features are vital for identifying conserved regions, current methods rely on computationally expensive multiple sequence alignment (MSA), creating a significant bottleneck. To address this, we propose MAFI-DTA, a novel model that circumvents the computational burden of MSA by deriving evolutionary context directly from sequences using the advanced protein language model, ESM-3. Furthermore, we introduced a multi-scale protein graph construction strategy based on varying numbers of residues, enabling the effective extraction of structural information across different scales through a neural network integrating graph neural networks, BiLSTM, and transformer modules. Experimental validation on multiple benchmark datasets demonstrates that MAFI-DTA achieves significantly better performance compared to existing approaches. This improved accuracy is attributed to the effective incorporation of both the predicted evolutionary information and the multi-scale graph representations. The source code is available at https://github.com/aliveadult/MAFI. MAFI-DTA successfully captures crucial evolutionary and multi-scale structural context efficiently, overcoming the limitations of traditional MSA methods. This approach provides a high-performing tool that facilitates the study of drug–target interactions and accelerates the drug discovery process.
Jiang et al. (Mon,) studied this question.