Multimodal deep learning with a joint uncertainty quantification scheme for drug-target interaction prediction

AI-driven prediction of drug-target interaction (DTI) has emerged as a critical component in modern drug discovery and development. However, this approach is constrained by model and data uncertainties, which substantially affect its reliability and accuracy in DTI prediction. To overcome these limitations, we introduce EUQTri-DTI, a novel evidence-guided uncertainty quantification-based multimodal deep learning framework for DTI prediction. Specifically, EUQTri-DTI is designed to integrate three modality-specific networks to encode 1D protein sequences, 2D molecular images, and 3D drug structures. Here, a bidirectional cross-attention mechanism is employed to facilitate information exchange between different modalities. Additionally, we develop a joint uncertainty quantification scheme by calculating the weight summation of evidential uncertainty and prediction entropy from the aforementioned output, enabling a more comprehensive and nuanced assessment of uncertainties. Experiments are presented to demonstrate that EUQTri-DTI achieves stable and competitive performance through multiple evaluation metrics on three benchmark datasets, compared to baseline and state-of-the-art methods. Specifically, EUQTri-DTI achieves ROC-AUC and PR-AUC values of 85.28% and 84.53% on DrugBank, 93.45% and 83.02% on KIBA, and 92.41% and 85.32% on Davis, respectively. Moreover, the uncertainty analyses indicate that misclassified samples generally exhibit higher uncertainty, supporting sample-level confidence assessment and reliability-aware decision making. Overall, EUQTri-DTI combines competitive predictive performance with sample-level reliability assessment and shows potential for uncertainty-aware virtual screening.