Tuberculosis (TB) remains one of the leading causes of death among infectious diseases, further complicated by the increasing prevalence of drug-resistant strains of Mycobacterium tuberculosis. One of the key enzymes in the biosynthesis of mycolic acids – the main components of the mycobacterial cell wall – is enoyl-ACP reductase InhA, making it an important target for the development of new antituberculosis agents.
In this master’s thesis, we designed an integrated approach combining bioinformatics analysis, machine learning, and experimental synthesis to identify and evaluate new inhibitors of the InhA enzyme. In the initial phase, we collected a dataset of more than 1400 known inhibitors of the ENR enzyme family (including InhA, FabI, FabV, and FabK) and categorized them as active or inactive based on their pIC50 values. Using KNIME and Python, we calculated a range of physicochemical descriptors – such as logP, the number of hydrogen bond donors and acceptors, topological polar surface area, and molecular weight – which served as additional criteria for subsequent compound filtering. The data were further analysed using t-SNE visualization, clustering, and structural fragment identification.
Through various machine learning methods and feature selection techniques (RFE, PCA, SHAP), we found that the presence of secondary amides, aromatic rings, and tautomerizable groups positively correlated with inhibitory activity, while a higher number of basic N-heterocycles and alkenes tended to reduce activity. Among the analysed compounds, oxopyrrolidine carboxamide, piperidine, and piperazine fragments emerged as key motifs.
Based on these findings, we designed a library of novel compounds using BROOD and additionally generated a virtual library in Python with the RDKit library. All compounds were subjected to virtual screening using Glide and ROCS, along with pharmacophore modelling. The developed predictive machine learning model was also applied to screen a commercial compound database to select the most promising candidates.
Following selection and molecular docking into the InhA active site, we identified five compounds for experimental synthesis. These were prepared using various chemical reactions, including Suzuki-Miyara coupling, carbodiimide-mediated condensation, and alkylation. The synthesized compounds were characterized by NMR, IR, LC-MS, and HRMS and submitted for testing against isolated InhA enzyme.
Although experimental results are not yet available, numerous literature sources indicate that the combination of computer-aided drug design and machine learning is highly promising for the discovery of new inhibitors. This work thus lays a solid foundation for further pharmacological and toxicological studies and the potential development of novel antituberculosis agents.
|
