Combining Stochastic Deformation/Relaxation and Intermolecular Contacts Analysis for Extracting Pharmacophores from Ligand-Receptor Complexes

Computational workflow for discovering small molecular binders for shallow binding sites by integrating molecular dynamics simulation, pharmacophore modeling, and machine learning: STAT3 as case study

STAT3 belongs to a family of seven transcription factors. It plays an important role in activating the transcription of various genes involved in a variety of cellular processes. High levels of STAT3 are detected in several types of cancer. Hence, STAT3 inhibition is considered a promising therapeutic anti-cancer strategy. However, since STAT3 inhibitors bind to the shallow SH2 domain of the protein, it is expected that hydration water molecules play significant role in ligand-binding complicating the discovery of potent binders. To remedy this issue, we herein propose to extract pharmacophores from molecular dynamics (MD) frames of a potent co-crystallized ligand complexed within STAT3 SH2 domain. Subsequently, we employ genetic function algorithm coupled with machine learning (GFA-ML) to explore the optimal combination of MD-derived pharmacophores that can account for the variations in bioactivity among a list of inhibitors. To enhance the dataset, the training and testing lists were augmented nearly a 100-fold by considering multiple conformers of the ligands. A single significant pharmacophore emerged after 188 ns of MD simulation to represent STAT3-ligand binding. Screening the National Cancer Institute (NCI) database with this model identified one low micromolar inhibitor most likely binds to the SH2 domain of STAT3 and inhibits this pathway.

Discovery of new potent lysine specific histone demythelase-1 inhibitors (LSD-1) using structure based and ligand based molecular modelling and machine learning

Lysine-specific histone demethylase 1 (LSD-1) is an epigenetic enzyme that oxidatively cleaves methyl groups from monomethyl and dimethyl Lys4 of histone H3 and is highly overexpressed in different types of cancer. Therefore, it has been widely recognized as a promising therapeutic target for cancer therapy. Towards this end, we employed various Computer Aided Drug Design (CADD) approaches including pharmacophore modelling and machine learning. Pharmacophores generated by structure-based (SB) (either crystallographic-based or docking-based) and ligand-based (LB) (either supervised or unsupervised) modelling methods were allowed to compete within the context of genetic algorithm/machine learning and were assessed by Shapley additive explanation values (SHAP) to end up with three successful pharmacophores that were used to screen the National Cancer Institute (NCI) database. Seventy-five NCI hits were tested for their LSD-1 inhibitory properties against neuroblastoma SH-SY5Y cells, pancreatic carcinoma Panc-1 cells, glioblastoma U-87 MG cells and in vitro enzymatic assay, culminating in 3 nanomolar LSD-1 inhibitors of novel chemotypes.

Discovery of new Cdc2-like kinase 4 (CLK4) inhibitors via pharmacophore exploration combined with flexible docking-based ligand/receptor contact fingerprints and machine learning

Cdc2-like kinase 4 (CLK4) inhibitors are of potential therapeutic value in many diseases particularly cancer. In this study, we combined extensive ligand-based pharmacophore exploration, ligand–receptor contact fingerprints generated by flexible docking, physicochemical descriptors and machine learning-quantitative structure–activity relationship (ML-QSAR) analysis to investigate the pharmacophoric/binding requirements for potent CLK4 antagonists. Several ML methods were attempted to tie these properties with anti-CLK4 bioactivities including multiple linear regression (MLR), random forests (RF), extreme gradient boosting (XGBoost), probabilistic neural network (PNN), and support vector regression (SVR). A genetic function algorithm (GFA) was combined with each method for feature selection. Eventually, GFA-SVR was found to produce the best self-consistent and predictive model. The model selected three pharmacophores, three ligand–receptor contacts and two physicochemical descriptors. The GFA-SVR model and associated pharmacophore models were used to screen the National Cancer Institute (NCI) structural database for novel CLK4 antagonists. Three potent hits were identified with the best one showing an anti-CLK4 IC₅₀ value of 57 nM.

Ligand-based pharmacophores, ligand–receptor contact fingerprints, physicochemical descriptors and machine learning were combined to probe binding of potent CLK4 antagonists. GFA-SVR gave the best model. Virtual screening identified 3 nanomolar hits.

Docking-generated multiple ligand poses for bootstrapping bioactivity classifying Machine Learning: Repurposing covalent inhibitors for COVID-19-related TMPRSS2 as case study

In the present work we introduce the use of multiple docked poses for bootstrapping machine learning-based QSAR modelling. Ligand-receptor contact fingerprints are implemented as descriptor variables. We implemented this method for the discovery of potential inhibitors of the serine protease enzyme TMPRSS2 involved the infectivity of coronaviruses. Several machine learners were scanned, however, Xgboost, support vector machines (SVM) and random forests (RF) were the best with testing set accuracies reaching 90%. Three potential hits were identified upon using the method to scan known untested FDA approved drugs against TMPRSS2. Subsequent molecular dynamics simulation and covalent docking supported the results of the new computational approach.

Structure-Based Discovery and Bioactivity Evaluation of Novel Aurora-A Kinase Inhibitors as Anticancer Agents via Docking-Based Comparative Intermolecular Contacts Analysis (dbCICA)

Aurora-A kinase plays a central role in mitosis, where aberrant activation contributes to cancer by promoting cell cycle progression, genomic instability, epithelial-mesenchymal transition, and cancer stemness. Aurora-A kinase inhibitors have shown encouraging results in clinical trials but have not gained Food and Drug Administration (FDA) approval. An innovative computational workflow named Docking-based Comparative Intermolecular Contacts Analysis (dbCICA) was applied—aiming to identify novel Aurora-A kinase inhibitors—using seventy-nine reported Aurora-A kinase inhibitors to specify the best possible docking settings needed to fit into the active-site binding pocket of Aurora-A kinase crystal structure, in a process that only potent ligands contact critical binding-site spots, distinct from those occupied by less-active ligands. Optimal dbCICA models were transformed into two corresponding pharmacophores. The optimal one, in capturing active hits and discarding inactive ones, validated by receiver operating characteristic analysis, was used as a virtual in-silico search query for screening new molecules from the National Cancer Institute database. A fluorescence resonance energy transfer (FRET)-based assay was used to assess the activity of captured molecules and five promising Aurora-A kinase inhibitors were identified. The activity was next validated using a cell culture anti-proliferative assay (MTT) and revealed a most potent lead 85(NCI 14040) molecule after 72 h of incubation, scoring IC₅₀ values of 3.5–11.0 μM against PANC1 (pancreas), PC-3 (prostate), T-47D and MDA-MB-231 (breast)cancer cells, and showing favorable safety profiles (27.5 μM IC₅₀ on fibroblasts). Our results provide new clues for further development of Aurora-A kinase inhibitors as anticancer molecules.

Comprehensive Structural and Molecular Comparison of Spike Proteins of SARS-CoV-2, SARS-CoV and MERS-CoV, and Their Interactions with ACE2

The severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) has recently emerged in China and caused a disease called coronavirus disease 2019 (COVID-19). The virus quickly spread around the world, causing a sustained global outbreak. Although SARS-CoV-2, and other coronaviruses, SARS-CoV and Middle East respiratory syndrome CoV (MERS-CoV) are highly similar genetically and at the protein production level, there are significant differences between them. Research has shown that the structural spike (S) protein plays an important role in the evolution and transmission of SARS-CoV-2. So far, studies have shown that various genes encoding primarily for elements of S protein undergo frequent mutation. We have performed an in-depth review of the literature covering the structural and mutational aspects of S protein in the context of SARS-CoV-2, and compared them with those of SARS-CoV and MERS-CoV. Our analytical approach consisted in an initial genome and transcriptome analysis, followed by primary, secondary and tertiary protein structure analysis. Additionally, we investigated the potential effects of these differences on the S protein binding and interactions to angiotensin-converting enzyme 2 (ACE2), and we established, after extensive analysis of previous research articles, that SARS-CoV-2 and SARS-CoV use different ends/regions in S protein receptor-binding motif (RBM) and different types of interactions for their chief binding with ACE2. These differences may have significant implications on pathogenesis, entry and ability to infect intermediate hosts for these coronaviruses. This review comprehensively addresses in detail the variations in S protein, its receptor-binding characteristics and detailed structural interactions, the process of cleavage involved in priming, as well as other differences between coronaviruses.

Discovery of New Phosphoinositide 3-kinase Delta (PI3Kδ) Inhibitors via Virtual Screening using Crystallography-derived Pharmacophore Modelling and QSAR Analysis

BACKGROUND

PI3Kδ is predominantly expressed in hematopoietic cells and participates in the activation of leukocytes. PI3Kδ inhibition is a promising approach for treating inflammatory diseases and leukocyte malignancies. Accordingly, we decided to model PI3Kδ binding.

METHODS

Seventeen PI3Kδ crystallographic complexes were used to extract 94 pharmacophore models. QSAR modelling was subsequently used to select the superior pharmacophore(s) that best explain bioactivity variation within a list of 79 diverse inhibitors (i.e., upon combination with other physicochemical descriptors).

RESULTS

The best QSAR model (r2 = 0.71, r2 LOO = 0.70, r2 press against external testing list of 15 compounds = 0.80) included a single crystallographic pharmacophore of optimal explanatory qualities. The resulting pharmacophore and QSAR model were used to screen the National Cancer Institute (NCI) database for new PI3Kδ inhibitors. Two hits showed low micromolar IC50 values.

CONCLUSION

Crystallography-based pharmacophores were successfully combined with QSAR analysis for the identification of novel PI3Kδ inhibitors.