TINDL

TINDL predicts anti-cancer drug response and identifies gene-expression biomarkers by applying an interpretable deep learning model with tissue-informed normalization between cancer cell lines (CCLs) and patient tumor samples.

Key Features:

• Tissue-Informed Normalization: Adjusts gene expression data by tissue type and cancer subtype to reduce discrepancies between preclinical CCLs and patient tumor data.
• Interpretable Deep Learning Model: Uses a deep learning architecture that yields a concise set of genes whose expression levels correlate with drug response, enabling biomarker identification.
• Comparative Performance: Evaluated across 14 drugs and distinguished sensitive versus resistant tumors for 10 drugs, outperforming other machine learning models in benchmark comparisons.
• Experimental Validation with siRNA: Predictions were experimentally validated using small interfering RNA (siRNA) knockdown; for tamoxifen, 10 genes were implicated in MCF7 cells and seven in T47D cells.
• Insights into Drug Mechanisms: Identifies genes associated with multiple drugs to reveal shared mechanisms of action and implicated signaling pathways.

Scientific Applications:

• Personalized treatment prediction: Predicts patient-level sensitivity and resistance to anti-cancer drugs using integrated CCL and tumor expression data.
• Biomarker discovery: Identifies gene-expression biomarkers that correlate with drug response for use in translational studies.
• Mechanistic investigation: Reveals genes and signaling pathways shared across multiple drugs to inform studies of mechanisms of action and therapeutic targets.

Methodology:

Trains an interpretable deep learning model on preclinical cancer cell line (CCL) data and tumor samples while applying tissue-informed normalization to align distributions and extract gene-expression biomarkers associated with drug response.