Mohan Rao is an accomplished drug discovery and development scientist with a Ph.D. and Post-Ph.D. in  Computational Biology/Chemistry from the Indian Institute of Science and The Scripps Research Institute,  La Jolla, respectively. His expertise spans diverse domains, including drug discovery, transcriptomics,  Machine Learning/Artificial Intelligence (AI/ML), High-Content Biosignatures data Analytics, Regulatory  Toxicology, Statistics, and Toxico & Bioinformatics. 

Currently serving as a Regulatory & Computational Toxicologist at Neurocrine Biosciences, he oversees vendor assessments, develops in vivo toxicology plans aligned with clinical development strategies for  CNS indications, and effectively utilizes computational toxicology to design and execute GLP and non-GLP in vivo studies for toxicity assessment. Previously, as an Associate Director of Predictive, Investigative, and  Translational Toxicology/Computational Toxicology at Janssen Pharmaceuticals, he developed computational systems toxicology, employed AI/ML approaches to predict DILI for small molecules, and provided investigative toxicology support across various therapeutic modalities. 

Before that, as the Computational Toxicology and chemistry Lead at AbbVie, he developed a novel off-target prediction framework for molecules of different sizes, contributing to the design and optimization of biologics and antibody-drug conjugate (ADC) platforms. 

In an earlier role, he served as the Director of Computational Drug Discovery at Transtech Pharma Inc., where he designed protein-protein interaction inhibitors for Alzheimer's disease, developed small molecule activators for GLP-1r, designed small molecule anticancer/anti-inflammatory compounds, and participated in the prediction of clinical biomarkers for various disease programs. 

He has presented research findings at various professional meetings and have published several peer-reviewed manuscripts and 25 US patents

OpenTox Virtual Conference 2023

AI/ML-driven Predictive Modeling of Drug-Induced Liver Injury (DILI) Severity in Small Molecules

Drug-induced liver injury (DILI) is a complex and multifactorial toxicity that has been a major cause of attrition in drug discovery, development, and post-marketing. Early identification of DILI risk is crucial for reducing the costs and time associated with drug development. While various predictive models based on physicochemical properties or in vitro and in vivo assays have been reported in recent years, these approaches often fail to account for the involvement of liver-expressed proteins in DILI. To bridge this gap, we have developed an integrated artificial intelligence/machine learning (AI/ML) model to predict the severity of DILI for small molecules. Our approach combines physicochemical properties with in silico-predicted off-target interactions to improve accuracy. Our dataset comprises 603 diverse compounds from public databases, classified by the FDA as Most DILI (M-DILI), Less DILI (L-DILI), and No DILI (N-DILI). We utilized six machine learning methods, namely k-nearest neighbor (k-NN), support vector machine  (SVM), random forest (RF), Naivë Bayes (NB), artificial neural network (ANN), logistic regression (LR),  weighted average ensemble learning (WA), and penalized logistic regression (PLR), to create a consensus model for predicting DILI potential. SVM, RF, LR, WA, and PLR demonstrated the best performance in identifying M-DILI and N-DILI compounds, with an receiver operating characteristic area under the curve of 0.88, sensitivity of 0.73, and specificity of 0.9. Through our analysis, we identified approximately 43 off-targets, in addition to essential physicochemical properties (fsp3, log S, basicity, reactive functional groups, and predicted metabolites), as critical factors in distinguishing between M-DILI and N-DILI  compounds. Notable key off-targets include PTGS1, PTGS2, SLC22A12, PPARγ, RXRA, CYP2C9, AKR1C3,  MGLL, RET, AR, and ABCC4.Our AI/ML computational approach shows the significance of integrating physicochemical properties with predicted on- and off-target biological interactions, substantially improving DILI predictivity compared to relying on chemical properties alone. This AI/ML model shows promise in identifying potential DILI risks early, enhancing the probability of success in clinic.