Biomarker Discovery

One of my major responsibilities at Accent was to identify candidate predictive biomarkers for its oncology drugs. Since I joined Accent, I developed a biomarker discovery platform capable of handling various tasks to identify and explore potential biomarkers. The platform consists of an Accent proprietary database and a set of code that can efficiently generate biomarker reports with compound response data.

The platform can do the following semi-automatic analysis:

Data input:

• IC50s/EC50s/AUCs for response-based analysis
• Sensitivity class for two class analysis
• Genomics data curated from CCLE/sanger/CROs

Routine biomarker discovery analysis at Accent:

• Damaging mutation enrichment (Fisher or Wilcoxon test)
• Copy number enrichment (limma::lmfit or Wilcoxon test)
• Gene expression (limma::lmfit or spearman correlation)
• Chronos and Demeter2 Scores (correlation)
• GDSC and PRISM response (correlation)
• Global markers, e.g. MSI, Aneuploidy, Whole genome Doubling, TMB (correlation or Wilcoxon)
• Feature ranking and selection (Random Forest, PLS-DA/PLS)
• Other (methylation, chromatin status, protein expression)

Output

• Interactive HTML report (Rmarkdown)