Seminarium Zakładu Biofizyki

The aim of this study is to develop and evaluate a comprehensive graph network, clustering analysis, and Machine Learning (ML) methods for classifying drugs based on their target types, utilizing their SMILES chemical structures. This approach is based on the hypothesis that highly similar compounds exhibit similar therapeutic effects and target types. Leveraging a DrugBank dataset containing chemical structures, biological activity data, and target information, we employed a multifaceted strategy integrating the construction and analysis of a comprehensive dataset. Within the graph network, high-centrality analysis effectively grouped molecules associated with the same target types. Additionally, hierarchical clustering successfully identified molecules with similar properties, exemplified by polar OH bonds observed in cluster 17. The ML model achieved an accuracy of 88%, underscoring the effectiveness of the proposed method, particularly given that only abbreviated 1-dimensional chemical structures were used as input features.