Intercepting Liver Toxicity Before Human Trials: AI Digital Liver Model Enters First Stage of FDA Qualification Review

Before a drug enters a Phase 1 human trial, the development team must answer a difficult question: Could a molecule that appears safe in animal and cell studies damage the liver in humans? The U.S. Food and Drug Administration (FDA) has accepted Absentia Labs’ letter of intent for its “digital liver model” into the ISTAND Drug Development Tool Qualification Program, marking the first time an artificial intelligence tool for predicting drug-induced liver injury has entered this formal evaluation pathway.

The tool targets small-molecule drug candidates that have not yet entered human trials. The FDA said the model compares the chemical structure of a new candidate with historical reference drugs whose liver-injury risks are known, helping assess potential hazards. The developer said the system also incorporates AI trained on mechanistic biology and drug-response data, with the goal of capturing liver toxicity arising from the interplay of factors including metabolism, immune responses, dose, and exposure.

Drug-induced liver injury is one of the major causes of clinical trial termination and new-drug development failure, while existing animal, in vitro, and computer models struggle to accurately reproduce human responses. The practical value of a digital model therefore lies not in issuing a “certificate of safety,” but in adding another piece of evidence to candidate-selection decisions: which molecules require additional testing, adjustments to dose design, or should not continue toward Phase 1 clinical trials.

The FDA expects that, if the tool is ultimately qualified, it will contribute alongside other nonclinical liver-safety data to an overall weight-of-evidence assessment, rather than independently determining whether a candidate can enter human trials. This positioning also means that the model’s ability to maintain reliable performance across different chemical scaffolds, mechanisms of action, and exposure conditions—and how its scope of application is defined—will be central to subsequent review.

The current progress represents only the first step of a three-stage process: the FDA’s acceptance of the letter of intent means the tool merits further evaluation; it does not mean qualification has been completed, clinical performance has been demonstrated, or regulatory endorsement has been obtained. Absentia must next submit a qualification plan and then a complete qualification package. Only after completing the entire process successfully may drug companies use the tool within the approved context of use.

The company also said its model was selected as one of the winners of the precisionFDA animal liver-toxicity prediction challenge and that it is developing other models of human biology. However, the competition results concerned animal liver-toxicity modeling and cannot directly demonstrate that the model can accurately predict drug-induced liver injury in humans. Publicly available information also does not disclose the scale of the training data, external-validation performance, or error rates. These evidence gaps are central to whether the digital liver can progress from a promising supplementary tool to a reusable regulatory tool.