pdCSM-cancer

pdCSM-cancer predicts anticancer activity and GI50% growth inhibition of small molecules using graph-based signature representations and machine learning to evaluate compounds across cancer cell lines.

Key Features:

• Graph-Based Signature Representation: Uses graph-based signature representations of chemical structures to encode molecular features for prediction.
• Extensive Training Data: Trained on over 18,000 compounds tested across 9 tumor types and 74 distinct cancer cell lines.
• Predictive Models: Includes trained and validated models that predict GI50% with cross-validation and independent test Pearson correlations up to 0.74 and comparable performance metrics up to 0.67.
• Generic Predictive Model: Provides a generic model to identify molecules active across at least 60 cancer cell lines with area under the ROC curve (AUC) up to 0.94 in cross-validation and independent tests.

Scientific Applications:

• Optimizing Screening Libraries: Prioritize and enrich screening libraries by predicting compounds' anticancer activity.
• Research and Development: Aid drug discovery and lead selection by informing molecular bioactivity and pharmacodynamics related to growth inhibition in cancer cell lines.

Methodology:

Analyzes graph-based chemical signatures using machine learning with trained and validated models; performance was assessed by cross-validation and independent tests using Pearson correlation coefficients and AUC metrics.