Exploring kinase family inhibitors and their moiety preferences using deep SHapley additive exPlanations.

BACKGROUND: While it has been known that human protein kinases mediate most signal transductions in cells and their dysfunction can result in inflammatory diseases and cancers, it remains a challenge to find effective kinase inhibitor as drugs for these diseases. One major challenge is the compensatory upregulation of related kinases following some critical kinase inhibition. To circumvent the compensatory effect, it is desirable to have inhibitors that inhibit all the kinases belonging to the same family, instead of targeting only a few kinases. However, finding inhibitors that target a whole kinase family is laborious and time consuming in wet lab. RESULTS: In this paper, we present a computational approach taking advantage of interpretable deep learning models to address this challenge. Specifically, we firstly collected 9,037 inhibitor bioassay results (with 3991 active and 5046 inactive pairs) for eight kinase families (including EGFR, Jak, GSK, CLK, PIM, PKD, Akt and PKG) from the ChEMBL25 Database and the Metz Kinase Profiling Data. We generated 238 binary moiety features for each inhibitor, and used the features as input to train eight deep neural networks (DNN) models to predict whether an inhibitor is active for each kinase family. We then employed the SHapley Additive exPlanations (SHAP) to analyze the importance of each moiety feature in each classification model, identifying moieties that are in the common kinase hinge sites across the eight kinase families, as well as moieties that are specific to some kinase families. We finally validated these identified moieties using experimental crystal structures to reveal their functional importance in kinase inhibition. CONCLUSION: With the SHAP methodology, we identified two common moieties for eight kinase families, 9 EGFR-specific moieties, and 6 Akt-specific moieties, that bear functional importance in kinase inhibition. Our result suggests that SHAP has the potential to help finding effective pan-kinase family inhibitors.

Identification of pan-kinase-family inhibitors using graph convolutional networks to reveal family-sensitive pre-moieties.

BACKGROUND: Human protein kinases, the key players in phosphoryl signal transduction, have been actively investigated as drug targets for complex diseases such as cancer, immune disorders, and Alzheimer's disease, with more than 60 successful drugs developed in the past 30 years. However, many of these single-kinase inhibitors show low efficacy and drug resistance has become an issue. Owing to the occurrence of highly conserved catalytic sites and shared signaling pathways within a kinase family, multi-target kinase inhibitors have attracted attention. RESULTS: To design and identify such pan-kinase family inhibitors (PKFIs), we proposed PKFI sets for eight families using 200,000 experimental bioactivity data points and applied a graph convolutional network (GCN) to build classification models. Furthermore, we identified and extracted family-sensitive (only present in a family) pre-moieties (parts of complete moieties) by utilizing a visualized explanation (i.e., where the model focuses on each input) method for deep learning, gradient-weighted class activation mapping (Grad-CAM). CONCLUSIONS: This study is the first to propose the PKFI sets, and our results point out and validate the power of GCN models in understanding the pre-moieties of PKFIs within and across different kinase families. Moreover, we highlight the discoverability of family-sensitive pre-moieties in PKFI identification and drug design.

Discovery of moiety preference by Shapley value in protein kinase family using random forest models.

BACKGROUND: Human protein kinases play important roles in cancers, are highly co-regulated by kinase families rather than a single kinase, and complementarily regulate signaling pathways. Even though there are > 100,000 protein kinase inhibitors, only 67 kinase drugs are currently approved by the Food and Drug Administration (FDA). RESULTS: In this study, we used "merged moiety-based interpretable features (MMIFs)," which merged four moiety-based compound features, including Checkmol fingerprint, PubChem fingerprint, rings in drugs, and in-house moieties as the input features for building random forest (RF) models. By using > 200,000 bioactivity test data, we classified inhibitors as kinase family inhibitors or non-inhibitors in the machine learning. The results showed that our RF models achieved good accuracy (> 0.8) for the 10 kinase families. In addition, we found kinase common and specific moieties across families using the Shapley Additive exPlanations (SHAP) approach. We also verified our results using protein kinase complex structures containing important interactions of the hinges, DFGs, or P-loops in the ATP pocket of active sites. CONCLUSIONS: In summary, we not only constructed highly accurate prediction models for predicting inhibitors of kinase families but also discovered common and specific inhibitor moieties between different kinase families, providing new opportunities for designing protein kinase inhibitors.

An integrated approach with new strategies for QSAR models and lead optimization.

BACKGROUND: Computational drug design approaches are important for shortening the time and reducing the cost for drug discovery and development. Among these methods, molecular docking and quantitative structure activity relationship (QSAR) play key roles for lead discovery and optimization. Here, we propose an integrated approach with core strategies to identify the protein-ligand hot spots for QSAR models and lead optimization. These core strategies are: 1) to generate both residue-based and atom-based interactions as the features; 2) to identify compound common and specific skeletons; and 3) to infer consensus features for QSAR models. RESULTS: We evaluated our methods and new strategies on building QSAR models of human acetylcholinesterase (huAChE). The leave-one-out cross validation values q 2 and r 2 of our huAChE QSAR model are 0.82 and 0.78, respectively. The experimental results show that the selected features (resides/atoms) are important for enzymatic functions and stabling the protein structure by forming key interactions (e.g., stack forces and hydrogen bonds) between huAChE and its inhibitors. Finally, we applied our methods to arthrobacter globiformis histamine oxidase (AGHO) which is correlated to heart failure and diabetic. CONCLUSIONS: Based on our AGHO QSAR model, we identified a new substrate verified by bioassay experiments for AGHO. These results show that our methods and new strategies can yield stable and high accuracy QSAR models. We believe that our methods and strategies are useful for discovering new leads and guiding lead optimization in drug discovery.

Methotrexate inhibition of SARS-CoV-2 entry, infection and inflammation revealed by bioinformatics approach and a hamster model.

Background: Drug repurposing is a fast and effective way to develop drugs for an emerging disease such as COVID-19. The main challenges of effective drug repurposing are the discoveries of the right therapeutic targets and the right drugs for combating the disease. Methods: Here, we present a systematic repurposing approach, combining Homopharma and hierarchal systems biology networks (HiSBiN), to predict 327 therapeutic targets and 21,233 drug-target interactions of 1,592 FDA drugs for COVID-19. Among these multi-target drugs, eight candidates (along with pimozide and valsartan) were tested and methotrexate was identified to affect 14 therapeutic targets suppressing SARS-CoV-2 entry, viral replication, and COVID-19 pathologies. Through the use of in vitro (EC50 = 0.4 µM) and in vivo models, we show that methotrexate is able to inhibit COVID-19 via multiple mechanisms. Results: Our in vitro studies illustrate that methotrexate can suppress SARS-CoV-2 entry and replication by targeting furin and DHFR of the host, respectively. Additionally, methotrexate inhibits all four SARS-CoV-2 variants of concern. In a Syrian hamster model for COVID-19, methotrexate reduced virus replication, inflammation in the infected lungs. By analysis of transcriptomic analysis of collected samples from hamster lung, we uncovered that neutrophil infiltration and the pathways of innate immune response, adaptive immune response and thrombosis are modulated in the treated animals. Conclusions: We demonstrate that this systematic repurposing approach is potentially useful to identify pharmaceutical targets, multi-target drugs and regulated pathways for a complex disease. Our findings indicate that methotrexate is established as a promising drug against SARS-CoV-2 variants and can be used to treat lung damage and inflammation in COVID-19, warranting future evaluation in clinical trials.

Pharmacophore anchor models of ATAT1 to discover potential inhibitors and lead optimization.

Post-translation modification of microtubules is associated with many diseases like cancer. Alpha Tubulin Acetyltransferase 1 (ATAT1) is a major enzyme that acetylates 'Lys-40' in alpha-tubulin on the luminal side of microtubules and is a drug target that lacks inhibitors. Here, we developed pharmacophore anchor models of ATAT1 which were constructed statistically using thousands of docked compounds, for drug design and investigating binding mechanisms. Our models infer the compound moiety preferences with the physico-chemical properties for the ATAT1 binding site. The results from the pharmacophore anchor models show the three main sub-pockets, including S1 acetyl site, S2 adenine site, and S3 diphosphate site with anchors, where conserved moieties interact with respective sub-pocket residues in each site and help in guiding inhibitor discovery. We validated these key anchors by analyzing 162 homologous protein sequences (>99 species) and over 10 structures with various bound ligands and mutations. Our results were consistent with previous works also providing new interesting insights. Our models applied in virtual screening predicted several ATAT1 potential inhibitors. We believe that our model is useful for future inhibitor discovery and for guiding lead optimization.

Uncovering Flexible Active Site Conformations of SARS-CoV-2 3CL Proteases through Protease Pharmacophore Clusters and COVID-19 Drug Repurposing.

The infectious SARS-CoV-2 causes COVID-19, which is now a global pandemic. Aiming for effective treatments, we focused on the key drug target, the viral 3C-like (3CL) protease. We modeled a big dataset with 42 SARS-CoV-2 3CL protease-ligand complex structures from ∼98.7% similar SARS-CoV 3CL protease with abundant complex structures. The diverse flexible active site conformations identified in the dataset were clustered into six protease pharmacophore clusters (PPCs). For the PPCs with distinct flexible protease active sites and diverse interaction environments, we identified pharmacophore anchor hotspots. A total of 11 "PPC consensus anchors" (a distinct set observed in each PPC) were observed, of which three "PPC core anchors" EHV2, HV1, and V3 are strongly conserved across PPCs. The six PPC cavities were then applied in virtual screening of 2122 FDA drugs for repurposing, using core anchor-derived "PPC scoring S" to yield seven drug candidates. Experimental testing by SARS-CoV-2 3CL protease inhibition assay and antiviral cytopathic effect assays discovered active hits, Boceprevir and Telaprevir (HCV drugs) and Nelfinavir (HIV drug). Specifically, Boceprevir showed strong protease inhibition with micromolar IC50 of 1.42 µM and an antiviral activity with EC50 of 49.89 µM, whereas Telaprevir showed moderate protease inhibition only with an IC50 of 11.47 µM. Nelfinavir solely showed antiviral activity with a micromolar EC50 value of 3.28 µM. Analysis of binding mechanisms of protease inhibitors revealed the role of PPC core anchors. Our PPCs revealed the flexible protease active site conformations, which successfully enabled drug repurposing.