Multimodal Fusion-Based\nLightweight Model for Enhanced\nGeneralization in Drug–Target Interaction Prediction

Predicting drug–target interactions (DTIs) with\nprecision\nis a crucial challenge in the quest for efficient and cost-effective\ndrug discovery. Existing DTI prediction models often require significant\ncomputational resources because of the intricate and exceptionally\nlengthy protein target sequences. This study introduces MMF-DTI, a\nlightweight model that uses multimodal fusion, to improve the generalizability\nof DTI predictions without sacrificing computational efficiency. The\nMMF-DTI model combines four distinct modalities: molecular sequence,\nmolecular properties, target sequence, and target function description.\nThis approach is noteworthy because it is the first to use natural\nlanguage-based target function descriptions in predicting DTIs. The\neffectiveness of MMF-DTI has been confirmed through benchmark data\nsets, demonstrating its comparable performance in terms of generalizability,\nespecially in scenarios with limited information about the drug or\ntarget. Remarkably, MMF-DTI accomplishes this using only half of the\nparameters and 17% of the VRAM compared with previous state-of-the-art\nmodels. This allows it to function even in constrained computational\nenvironments. The combination of performance and efficiency highlights\nthe potential of multimodal data fusion in improving the overall applicability\nof models, providing promising opportunities for future drug discovery\nendeavors.