Generating new and efficient drugs through machine learning-assisted quantitative structure–activity relationships (ML-QSAR) has become a promising strategy for addressing multidrug-resistant gram-negative bacterial infections. We developed robust ML-QSAR models using Genetic Function Approximation (GFA), Support Vector Machine (SVM), and Artificial Neural Network (ANN) methods to predict the activity of experimentally known quinoline-based MsbA inhibitors, aiming to create more effective antibacterial drugs. The ML models were built using eight significant molecular descriptors: B09[N-Cl], CATS3D_04_AA, F06[N-O], G2i, molecular weight (MW), Mor04p, VE1sign_B(s), and VE1sign_Dz(i), along with 279 molecular fingerprints to predict the MsbA inhibition activity of quinoline-based compounds. The molecular descriptor-based SVM model achieved an R² of 0.9891 and a q² cross-validation correlation of 0.9388. In contrast, the molecular fingerprint-based SVM model had an R² of 0.9981 and a q² cross-validation correlation of 0.7584, making it the best-performing model. The robustness of these developed models was further validated through various internal, external, and applicability domain analyses. The most active compounds identified in this data set, compounds 31 and 40, were subsequently used to generate 62 new quinoline-based compounds. Additionally, three modelled quinoline-based inhibitors, M28, N7, and N23, demonstrated excellent bioactivity, binding affinity, and pharmacokinetic profiles. To support further research, the fingerprint-based ML-QSAR model is available as a web application, MsbA-Pred (https://msba-mohan-amrita.streamlit.app/), which allows users to predict MsbA inhibitory activity from any device.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号