Interactome-based deep learning and path-level decomposition reveal TGFβ-IFNα competition in MASH

An interactome-based deep learning ensemble trained on drug-induced transcriptomics from primary human hepatocytes identifies competitive modulation between TGFβ and IFN signaling in MASH, via a four-node SMAD3–TP63–STAT2–STAT3 module.

• Why it matters: Dysregulated pro- and anti-fibrotic signaling crosstalk drives MASH fibrosis, but the mechanism and whether it acts directly on hepatocytes were unclear.
• What they did: DT-LEMBAS trained an ensemble of 50 models on simulated receptor knockouts in primary human hepatocytes, then used Jacobian-based chain-rule decomposition to pinpoint SMAD3, TP63, STAT2, and STAT3.
• The result: Gradient-derived mechanistic hypotheses were experimentally validated in primary hepatocyte spheroids, enabling interpretable decoding of large signaling neural networks into biological subnetworks.