In this talk, we discuss the paper “TwinCell: Large Causal Cell Model for Reliable and Interpretable Therapeutic Target Prioritisation” by J.-B. Morlot et al., bioarxiv, 2026.<\/p>\n
Abstract<\/strong><\/p>\n Drug discovery is impeded by the difficulty of translating targets from preclinical models to patients. Here, we present TwinCell, a Large Causal Cell Model (LCCM) capable of generalising from\u00a0in vitro<\/em>\u00a0cell lines to patient-derived cell types while providing biologically meaningful interpretations of its predictions. Rather than predicting perturbation outcomes, TwinCell focuses on identifying the upstream regulators driving the transition between diseased and healthy states. By integrating single-cell foundation model embeddings with a multiomics interactome, TwinCell constrains predictions to mechanistically plausible signalling routes. To validate this approach, we introduce TwinBench, a novel benchmarking framework to assess virtual cell models generalisation capability while correcting for popularity bias and mode collapse through empirical p-value estimation. When applied to\u00a0in clinico<\/em>\u00a0data, TwinCell recovers known drug targets aligned with the biology of the disease and demonstrates generalisation performance on targets unseen during training. This highlights the model\u2019s ability to learn mechanistic principles in a biological context. TwinCell represents a significant step toward building reliable and interpretable virtual cells for target identification bridging the gap between high-throughput\u00a0in vitro<\/em>\u00a0experiments and clinical insights.<\/p>\n","protected":false},"excerpt":{"rendered":" In this talk, we discuss the paper “TwinCell: Large Causal Cell Model for Reliable and Interpretable Therapeutic Target Prioritisation” by J.-B. Morlot et al., bioarxiv, 2026. Abstract Drug discovery is … <\/p>\n