Drugs_AI

The dataset used in this study consists of 9,378 chemical compounds represented through molecular descriptor profiles and biological activity measurements. Initially, the dataset contained 123 features, including compound identifiers, SMILES representations, physicochemical descriptors, topological indices, and MQN-based molecular characteristics. The target variable was identified as ic50_effect_size, which represents the biological activity level of each compound and serves as the prediction objective in the machine learning framework.

The dataset includes chemically diverse flavonoid compounds and their derivatives, such as kaempferol, apigenin, and quercetin, providing structural variability suitable for predictive modeling and quantitative structure–activity relationship (QSAR) analysis. Molecular descriptors were generated from SMILES (Simplified Molecular Input Line Entry System) representations, which encode the structural composition of chemical molecules in a machine-readable format. These descriptors capture multiple physicochemical and structural properties, including molecular weight, lipophilicity, hydrogen bonding capacity, ring structures, surface area descriptors, and topological characteristics.