Abstract
Preclinical models are used extensively to study diseases and therapies. In vitro monoculture or microphysiological system (MPS) platforms incorporating multiple different human cell types can emulate diseased tissues, but determining experimental conditions (e.g., media supplements) that provide the most effective translatability to humans (in vivo) is a major challenge. Using metabolic dysfunction–associated steatotic liver disease (MASLD) as a case study, we developed a machine learning framework [called LIV2TRANS (Latent In Vitro to In Vivo Translation)] that first maps MPS onto in vivo data, then elucidates translation insights, and lastly nominates experimental conditions that increase translatability. Our findings highlight TGFβ (transforming growth factor–β) as a crucial cue for MPS translatability and indicate that adding interferon-mediated JAK (Janus kinase)-STAT (signal transducer and activator of transcription) signaling perturbations could increase the predictive performance of MPS for MASLD. Last, an optimization algorithm highlights key signaling pathways to maximize germane human-relevant information captured by this MPS. This work establishes a mathematically principled approach for identifying experimental conditions that most beneficially capture in vivo–relevant molecular processes, generalizable to a wide range of diseases where suitable molecular data exist.