همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Using interpretable deep learning models and multi-objective data for computational discovery of new drugs

نویسندگان :

Masoud Ahmadlou¹

1- دانشگاه تحصیلات تکمیلی علوم پایه زنجان

کلمات کلیدی :

Computational Drug Discovery،Interpretable Deep Learning،Multimodal Data،Drug-Target Interaction،Bioinformatics

چکیده :

Abstract In recent years, deep learning (DL) has emerged as a pivotal approach in computational drug discovery. Despite its promise, challenges such as the interpretability of DL models and the underutilization of multimodal data remain unresolved. This study introduces an innovative approach that combines interpretable deep learning models with multimodal datasets to enhance drug discovery processes. Our methodology integrates biological data (e.g., protein sequences, pathways, and protein-protein interactions), chemical data (e.g., molecular structures), and clinical data (e.g., trial outcomes and side effects) sourced from PubChem, UniProt, and KEGG databases. The proposed model employs hybrid architectures, including graph neural networks (GNNs) and attention-based mechanisms, to predict drug-target interactions effectively. The framework excels not only in predictive accuracy but also in providing biologically and chemically interpretable insights into the underlying mechanisms of action. Experimental evaluations on benchmark datasets demonstrate that our method significantly outperforms state-of-the-art techniques in identifying novel drug candidates. Furthermore, the inclusion of interpretability enhances the understanding of drug interactions, making the model a valuable tool for reducing the time and cost associated with drug discovery. This framework contributes to addressing longstanding challenges in the field and facilitates more informed decision-making in pharmaceutical research.