De-risk targets through mechanism-enabled simulations in the Virtual Lab

Minimizing the risk of late-stage failure in drug discovery for novel targets remains a central challenge in translational research. Conventional preclinical perturbation datasets lack mechanistic insight and context specificity. Turbine's Virtual Lab offers access to simulations with mechanistic insights enabled by AI-guided virtual cell modeling to predict phenotypic and transcriptomic outcomes of genetic and pharmacological perturbations in preclinical models representative of patient heterogeneity. This framework enables systematic evaluation of target-disease linkage mechanisms, target liabilities, and combinatorial strategies to enrich preclinical decision-making data packages. We utilized our Simulated Cell technology to construct virtual replicates of more than 1,200 cancer cell lines by integrating large-scale omics data (CCLE) with a curated signaling network as prior knowledge. The virtual cell lines were trained on experimentally observed genetic (DepMap) and pharmacological perturbation data (GDSC2) to accurately represent their phenotypic responses, and on differential gene expression data (LINCS) to describe post-perturbation transcriptomic states. Benchmarking was performed against an unseen set of cell lines across all genes and selective essential genes only. Target assessment reports were generated by merging simulation-derived phenotypic and post-perturbation omics features with curated clinical intelligence, yielding an integrated view of signaling-level mechanisms of action, patient stratification strategies, combination potential, and initial safety assessments. The Simulated Cells accurately reproduced DepMap dependency outcomes, demonstrating a global test-set Pearson correlation of 0.90 across all genes. For selective essential genes only, we measured a cell-line wise 0.55 Pearson correlation. Replicate-level Pearson correlation is 0.78 for the same metric, while baseline (bias) models score only 0.23. Notably, the prospective validation hit rate stands at 70% for dependency predictions, also capturing novel dependencies not identified in DepMap. For a set of 10 in silico-identified targets, we observed a 50% shorter timeline from identification to in vivo validation and a 100% higher rate of in vitro validation compared with industry benchmarks. We have also established a Virtual Lab platform that provides easy, scalable, and interpretable access to these simulations. Turbine's target discovery platform combines mechanistic interpretability with strong concordance to experimental dependencies, bridging computational modeling and translational research. The Virtual Lab interface is user-friendly and optimized for adoption without any data science or AI expertise. This approach enables data-driven, de-risked, and accelerated target selection and portfolio advancement in oncology discovery.