Predicting Drug-Induced Liver Injury with Bayesian Machine Learning

Abstract

Drug induced liver injury (DILI) can require significant risk management in drug development and on occasion can cause morbidity or mortality, leading to drug attrition. Optimizing candidates preclinically can minimize hepatotoxicity risk, but it is difficult to predict due to multiple etiologies encompassing DILI, often with multifactorial and overlapping mechanisms. In addition to epidemiological risk factors, physicochemical properties, dose, disposition, lipophilicity, and hepatic metabolic function are also relevant for DILI risk. Better human-relevant, predictive models are required to improve hepatotoxicity risk assessment in drug discovery. Our hypothesis is that integrating mechanistically relevant hepatic safety assays with Bayesian machine learning will improve hepatic safety risk prediction. We present a quantitative and mechanistic risk assessment for candidate nomination using data from in vitro assays (hepatic spheroids, BSEP, mitochondrial toxicity, and bioactivation), together with physicochemical (cLogP) and exposure (Cmax total ) variables from a chemically diverse compound set (33 no/ low-, 40 medium-, and 23 high-severity DILI compounds). The Bayesian model predicts the continuous underlying DILI severity and uses a data-driven prior distribution over the parameters to prevent overfitting. The model quantifies the probability that a compound falls into either no/low-, medium-, or high-severity categories, with a balanced accuracy of 63% on held-out samples, and a continuous prediction of DILI severity along with uncertainty in the prediction. For a binary yes/no DILI prediction, the model has a balanced accuracy of 86%, a sensitivity of 87%, a specificity of 85%, a positive predictive value of 92%, and a negative predictive value of 78%. Combining physiologically relevant assays, improved alignment with FDA recommendations, and optimal statistical integration of assay data leads to improved DILI risk prediction.

A Next-Generation Risk Assessment Case Study for Coumarin in Cosmetic Products

“…Using historical benchmark chemicals with well-characterized human exposures and known human outcomes could help define a protective exposure level and an evidence-based MoS values. This benchmarking approach has not yet been developed for overall systemic toxicity, but examples applied to drug-induced liver injury ( Albrecht et al , 2019 ; Williams et al , 2020 ), reproductive ( Becker et al , 2015 ; Dent et al , 2018b ), and cardiac toxicity ( Lazic et al , 2018 ) have been previously published.…”

A compound attributes-based predictive model for drug induced liver injury in humans

“…Accurate and timely prediction of drug induced liver injury remains a challenging research topic with a great potential impact in drug R&D. A number of efforts have been made to build in silico models that can predict DILI [32][33][34]. Since the datasets and feature selection vary, a direct comparison between these methods and our approach may be inaccurate.…”

The evolution of strategies to minimise the risk of human drug-induced liver injury (DILI) in drug discovery and development

“…Supplemental Table 1 contains 1383 DILI and non-DILI reference compounds utilised from 19 publications (Aleo et al 2019;Albrecht et al 2019;Chen et al 2011Chen et al , 2016Dawson et al 2012;Garside et al 2014;Gustafsson et al 2014;Khetani et al 2012;O'Brien et al 2006;Porceddu et al 2012;Proctor et al 2017;Sakatis et al 2012;Schadt et al 2015;Tolosa et al 2012Tolosa et al , 2019Williams et al 2019). If available and/or interpretable the in vivo DILI observation has been simplified to positive (+ ve) or negative (−ve), and the assay or strategy prediction included, 422 compounds are classified in the LTKB, however, only 189 of these compounds are consistently classified as either DILI negative or positive across the publications.…”

“…To date, only limited studies have combined multi-parametric analysis combined with 3D approaches (e.g. Williams et al 2019). Often only s s e L e t a n o r d n e l a chlorpromazine Less -VE +VE +VE +VE +VE -VE +VE -VE +VE +VE +VE +VE +VE +VE +VE +VE +VE cyclophosphamide Less -VE +VE - --e n i l y g r a p Compounds: 2 3 2 5 2 6 2 3 1 8 3 0 37 3 7 3 7 37 4 9 49 49 49 29 54 54 39 39% 76% 88% 48% 89% 67% 59% 81% 54% 78% 55% 61% 67% 76% 93% 91% 93% 90% biochemical endpoints such as ATP have been adopted.…”