Synthesis of quinoline carboxamide analogues and comprehensive evaluation of their biological activities using bioactivity aware hybrid machine learning framework

Abstract

A series of six quinoline–carboxamide conjugates (2a–2f) prepared and screened at various concentrations for their anti-diabetic and anti-inflammatory effects in this research. The synthesis involved coupling quinoline carboxylic acid derivatives with several amines by means of O-(Benzotriazol-1-yl)-N, N, N', N'-tetramethyluronium tetrafluoroborate (TBTU) mediated amidation. Structural characterization is confirmed by 1H and 13C- NMR, FT-IR, UV-Vis, and ESI-MS. The biological activities are assessed using BSA denaturation and α-amylase inhibition assays. A Residual Network-50-based Stochastic Multi-Objective Optimized Deep Neural Network (ResNet -SMODNN) with multitask learning developed to predict the anti-diabetic and anti-inflammatory activities of quinoline–carboxamide conjugates (2a–2f). The model developed with the concentration dependent parameters and molecular descriptors such as molecular weight to explain the relationship among the compound properties and biological activity. It demonstrated excellent predictive performance with a coefficient of determination, MSE, MAE and RMSE of 1.00, 0.000, 0.011, 0.01 for training and 0.86, 0.056, 0.189 and 0.24 for testing, respectively. Among the synthesized derivatives, compound 2f had the best observed activity with experimentally determined IC₅₀ values of 6.35 µg/mL and 5.75 µg/mL for anti-diabetic and anti-inflammatory activities, respectively. According to SHAP analysis, the contribution of molecular weight the largest for the input features considered in the explored dataset. The hybrid machine learning approach gives a preliminary computational method for the bioactivity prioritization of quinoline–carboxamide derivatives and help in future experimental studies.