CEVIChE

CEVIChE predicts cellular viability from transcriptomics data by integrating LINCS-L1000 expression signatures with cell viability measurements to model phenotypes and derive consensus compound signatures for cell-specific drug sensitivity prediction.

Key Features:

• Data integration: Integrates LINCS-L1000 transcriptomics with cell viability data from the Achilles project and CTRP screen, producing over 90,000 signature–viability pairs.
• Phenotype modeling: Models cellular phenotypes from gene expression signatures to relate perturbations to viability outcomes.
• Viability prediction: Predicts cellular viability directly from transcriptomics signatures.
• Cell viability signature discovery: Identifies that cell viability signatures are a major component underlying perturbation signatures.
• Transcription factor linkage: Links perturbation signatures to transcription factors regulating cell death, proliferation, and division time.
• Consensus compound signatures: Derives consensus compound signatures that predict cell-specific drug sensitivity even when original signatures were not measured in the same cell line.
• Compound identification and validation: Identifies and validates compounds capable of inducing cell death in tumor cell lines.
• Toxicity confounding analysis: Addresses the confounding effect of cellular toxicity on signature similarity that can obscure mechanism-of-action discovery.
• Comparative predictive power: Shows consensus signatures predict drug sensitivity more effectively than conventional drug-specific features.

Scientific Applications:

• Drug sensitivity prediction: Predicts cell-specific drug sensitivity from transcriptomic signatures and consensus compound signatures.
• Mechanism-of-action deconvolution: Helps separate toxicity-driven signature similarity from mechanism-of-action-specific signals.
• Compound prioritization for tumor killing: Enables identification and validation of compounds that induce cell death in tumor cell lines.
• Functional genomics: Links transcription factor activity to viability-related transcriptional programs such as cell death, proliferation, and division time.
• Phenotypic modeling for drug discovery: Supports modeling of phenotypic consequences of genetic and chemical perturbations from expression data.

Methodology:

Integrates LINCS-L1000 transcriptomics with Achilles and CTRP screen viability measurements to generate >90,000 signature–viability pairs, models cellular phenotypes from expression to predict viability, derives consensus compound signatures, and analyzes transcription factor associations and toxicity-driven signature similarity confounding.