Overexpression of myeloid cell leukaemia-1 (Mcl-1) is commonly observed in numerous haematological malignancies and solid tumours, where it contributes to tumour progression, poor prognosis, and resistance to chemotherapy. Previous computational studies on Mcl-1 inhibitors (Mcl-1i) have predominantly focused on limited datasets with single scaffold, few modelling strategies, or structure-based analyses. In contrast, the present work presents a large-scale chemical space exploration integrated with fragment/fingerprint-based analysis and multiple machine learning approaches applied to a diverse Mcl-1i dataset. The current study characterizes scaffold diversity and Mcl-1 inhibitory activity trends within a large chemical space. Fragment-based QSAR modelling was implemented using Bayesian classification, thereby enabling the identification of both favourable and unfavourable fingerprints that regulate Mcl-1 inhibition. Furthermore, the significance of the identified fingerprints was validated through protein-ligand interaction analysis. In addition, statistically validated AI/ML models were developed that enable researchers to efficiently screen potential Mcl-1i. Positive fingerprints such as G9, G17, and G20 are consistently associated with low-nanomolar IC50 values, indicating that features combining aromaticity, flexible linkers, and heteroaromatic scaffolds strongly promote high-affinity binding. All these fingerprints as a collective offer mechanistically interpretable information on structure-activity correlations and act as useful leads to rational design and lead optimization of potent Mcl-1i.
Goswami et al. (Thu,) studied this question.