Potential dual inhibitors of Hexokinases and mitochondrial complex I discovered through machine learning approach

Abstract

Hexokinases (Hks) and mitochondrial complex I (MCI) are involved in the energy metabolism of cells; glycolysis/fermentation and oxidative phosphorylation. Both Hks and MCI are known to play critical roles in either division of metabolic plasticity which enables tumor progression and proliferation in the presence of chemotherapies. Therefore, targeting these enzymes are important in cancer drug resistance. Here, computational models for the prediction of inhibition of Hks were developed based on experimental data and an optimal feature subset that was selected by the Boruta algorithm (a wrapper feature selection algorithm coupled with random forest). Out of the seven models that were explored, a random forest classifier gave the best prediction (GA = 0.84, FNR = 0.12 and AUC = 0.96 for the external dataset). Fragmentation analysis led to the identification of the unique structural scaffolds that characterize hexokinase inhibitors and non-inhibitors. The best Hks inhibition model predicted that 23 molecules out of the 191 dataset of MCI actives (IC50 ≤ 10 µM) that were screened, have more than 60 % probability of exhibiting Hk inhibitory activity. Hence, they are possible dual inhibitors of both targets. Furthermore, the 23 molecules’ core structures are members of the scaffolds that are unique to Hk inhibitors earlier predicted by fragment analysis. The need for dual targeting agents in cancer therapy, particularly in combating cancer drug resistance, highlights the relevance of these findings.

Keywords

• Hexokinases
• Mitochondrial complex I
• Cancer
• MACCS fingerprints
• Boruta algorithms
• Machine learning