BCheM: a framework integrating molecular attributes and behavioral representations for precise prediction of drug resistance across multiple disease contexts.

Introduction: In clinical medicine, the response of diseases to drugs directly affects treatment efficacy and patient prognosis. Accurate prediction of drug responses can facilitate personalized treatment planning, enable the selection of effective medications, and reduce adverse effects. However, existing computational methods are often limited to single disease types or rely solely on molecularlevel information, making it difficult to simultaneously capture the semantic attribute features of diseases and drugs, as well as their behavioral patterns within association networks. Therefore, effectively integrating multimodal attribute information with complex relational structures for drug resistance prediction across multiple disease contexts remains a challenging problem. Methods: To address this issue, we propose a novel disease-drug resistance prediction framework, BCheM. First, a disease-drug resistance network is constructed based on experimentally validated drug response data, and rich attribute features are obtained by incorporating drug SMILES representations and disease semantic descriptions. Then, pre-trained models are employed to encode disease semantics and drug molecular structures, enabling deep representation of molecular attributes. Furthermore, a graph representation learning model that integrates Chebyshev graph convolution with a multi-head attention mechanism is developed to model and fuse behavioral features within the disease-drug network. Finally, a classifier is applied to predict potential disease-drug resistance associations. Results: Experimental results and case studies demonstrate that BCheM achieves superior performance across multiple evaluation metrics, exhibiting strong predictive capability and practical applicability. Discussion: This study provides an effective graph-based modeling approach for drug response prediction in multi-disease contexts.