BAY1128688 is a selective inhibitor of AKR1C3, investigated recently in a trial that was prematurely terminated due to drug-induced liver injury. These unexpected observations prompted use of the quantitative systems toxicology model, DILIsym, to determine possible mechanisms of hepatotoxicity. Using mechanistic in vitro toxicity data as well as clinical exposure data, DILIsym predicted the potential for BAY1128688 to cause liver toxicity (elevations in serum alanine aminotransferase (ALT)) and elevations in serum bilirubin. Initial simulations overpredicted hepatotoxicity and bilirubin elevations, so the BAY1128688 representation within DILIsym underwent optimization. The liver partition coefficient Kp was altered to align simulated bilirubin elevations with those observed clinically. Altering the mode of bile acid canalicular and basolateral efflux inhibition was necessary to accurately predict ALT elevations. Optimization results support that bilirubin elevations observed early during treatment are due to altered bilirubin metabolism and transporter inhibition, which is independent of liver injury. The modeling further supports that on-treatment ALT elevations result from inhibition of bile acid transporters, particularly the bile salt excretory pump, leading to accumulation of toxic bile acids. The predicted dose-dependent intrinsic hepatotoxicity may increase patient susceptibility to an adaptive immune response, accounting for ALT elevations observed after completion of treatment. These BAY1128688 simulations provide insight into the mechanisms behind hepatotoxicity and bilirubin elevations and may inform the potential risk posed by future compounds.

Drug discovery and development is a notoriously risky process with high failure rates at every stage, including disease modeling, target discovery, hit discovery, lead optimization, preclinical development, human safety, and efficacy studies. Accurate prediction of clinical trial outcomes may help significantly improve the efficiency of this process by prioritizing therapeutic programs that are more likely to succeed in clinical trials and ultimately benefit patients. Here, we describe inClinico, a transformer-based artificial intelligence software platform designed to predict the outcome of phase II clinical trials. The platform combines an ensemble of clinical trial outcome prediction engines that leverage generative artificial intelligence and multimodal data, including omics, text, clinical trial design, and small molecule properties. inClinico was validated in retrospective, quasi-prospective, and prospective validation studies internally and with pharmaceutical companies and financial institutions. The platform achieved 0.88 receiver operating characteristic area under the curve in predicting the phase II to phase III transition on a quasi-prospective validation dataset. The first prospective predictions were made and placed on date-stamped preprint servers in 2016. To validate our model in a real-world setting, we published forecasted outcomes for several phase II clinical trials achieving 79% accuracy for the trials that have read out. We also present an investment application of inClinico using date stamped virtual trading portfolio demonstrating 35% 9-month return on investment.

Infants with neonatal opioid withdrawal syndrome commonly receive morphine treatment to manage their withdrawal signs. However, the effectiveness of this pharmacotherapy in managing the infants' withdrawal signs vary widely. We sought to understand how information available early in infant monitoring can anticipate this treatment response, focusing on early modified Finnegan Neonatal Abstinence Scoring System (FNASS) scores, polygenic risk for opioid dependence (polygenic risk score (PRS)), and drug exposure. Using k-means clustering, we divided the 213 infants in our cohort into 3 groups based on their FNASS scores in the 12 hours before and after the initiation of pharmacotherapy. We found that these groups were pairwise significantly different for risk factors, including methadone exposure, and for in-hospital outcomes, including total morphine received, length of stay, and highest FNASS score. Whereas PRS was not predictive of receipt of treatment, PRS was pairwise significantly different between a subset of the groups. Using tree-based machine learning methods, we then constructed network graphs of the relationships among these groups, FNASS scores, PRS, drug exposures, and in-hospital outcomes. The resulting networks also showed meaningful connection between early FNASS scores and PRS, as well as between both of those and later in-hospital outcomes. These analyses present clinicians with the opportunity to better anticipate infant withdrawal progression and prepare accordingly, whether with expedited morphine treatment or non-pharmacotherapeutic alternative treatments.

In clinical trials of cannabidiol (CBD) for the treatment of seizures in patients with Dravet syndrome, Lennox-Gastaut syndrome, and tuberous sclerosis complex, elevations in serum alanine aminotransferase (ALT) > 3× the upper limit of normal were observed in some patents, but the incidence was much greater in patients who were receiving treatment with valproate (VPA) before starting CBD. To explore potential mechanisms underlying this interaction, we used DILIsym, a quantitative systems toxicology model, to predict ALT elevations in a simulated human population treated with CBD alone, VPA alone, and when CBD dosing was starting during treatment with VPA. We gathered in vitro data assessing the potential for CBD, the two major CBD metabolites, and VPA to cause hepatotoxicity via inhibition of bile acid transporters, mitochondrial dysfunction, and production of reactive oxygen species (ROS). Physiologically-based pharmacokinetic models for CBD and VPA were used to predict liver exposure. DILIsym simulations predicted dose-dependent ALT elevations from CBD treatment and this was predominantly driven by ROS production from the parent molecule. DILIsym also predicted VPA treatment to cause ALT elevations which were transient when mitochondrial biogenesis was incorporated into the model. Contrary to the clinical experience, simulation of 2 weeks treatment with VPA prior to introduction of CBD treatment did not predict an increase of the incidence of ALT elevations relative to CBD treatment alone. We conclude that the marked increased incidence of CBD-associated ALT elevations in patients already receiving VPA is unlikely to involve the three major mechanisms of direct hepatotoxicity.