RCSM

RCSM evaluates connectivity-scoring methods for L1000-based Connectivity Map (CMap) gene-expression data to quantify perturbagen-induced similarities for mechanism-of-action inference and drug repurposing.

Key Features:

• Six published methods evaluated: Systematic comparison of six established connectivity-scoring methods on L1000 data.
• Benchmark dataset: Uses drug-drug similarities from the Drug Repurposing Hub database as the benchmark standard.
• Performance metric: Assesses predictive accuracy using partial area under the receiver operating characteristic curve at false positive rates AUC0.001, AUC0.005, and AUC0.01.
• Signature size comparison: Evaluates method performance across gene signature sizes ranging from 10 to 200 genes.
• Top-performing method identified: Reports ZhangScore as exhibiting superior accuracy across evaluated signature sizes.
• Experimental-signature testing: Validates method performance on experimentally derived signatures, including estrogen response in breast cancer cells.
• Target-inhibitor discovery: Applies scoring results to identify candidate TOP2A inhibitors from TOP2A knockdown signatures.
• R implementation: Implements the six connectivity methods in an R package.

Scientific Applications:

• Mechanism-of-action inference: Quantifies similarities between perturbagen-induced gene-expression profiles to support MoA elucidation.
• Drug repurposing: Prioritizes drug-drug relationships for identifying new therapeutic indications using L1000 CMap data.
• Benchmarking connectivity metrics: Compares connectivity-scoring methods to guide selection of accurate scoring approaches for L1000 analyses.
• Target inhibitor identification: Identifies and prioritizes candidate inhibitors, exemplified by discovery efforts for TOP2A.
• Experimental signature evaluation: Tests robustness of scoring methods on experimentally derived signatures such as estrogen responses in breast cancer cells.

Methodology:

Systematic evaluation of six published connectivity scoring methods; benchmarking against drug-drug similarities from the Drug Repurposing Hub; performance assessed using partial AUC of ROC at false positive rates 0.001, 0.005, and 0.01; comparisons across gene signature sizes of 10–200 genes; applications to experimentally derived signatures (estrogen in breast cancer) and TOP2A knockdown signatures; methods implemented in an R package.