Indication prioritization & target validation

Use K-Pro to assess whether a target is worth advancing, and to check that biomarkers behave consistently across data modalities.

• Assess target tractability & druggability: evaluate target tractability, safety profiles, and therapeutic potential using curated knowledge bases (such as Open Targets, ChEMBL, DepMap, and the Human Protein Atlas).
• Validate biomarkers across modalities: test biomarkers and hypotheses across bulk RNA-seq, mutations, proteomics, single-cell, spatial, and clinical data.

Find candidates with convergent evidence

What to do: Combine several data modalities (for example RNA-seq and mutation data) within a cohort to find candidates supported by more than one layer of evidence. Start with a specific comparison between two subgroups rather than "analyze everything": focused questions give clearer, faster results.

Example prompt:

In the TCGA-LUAD cohort, integrate RNA-seq gene expression and mutation data. Identify genes that are both differentially expressed and frequently mutated in current smokers versus never-smokers.

Expected result: A differential expression analysis (volcano plot) and a mutation landscape (oncoprint) for the two subgroups, with a summary of the genes that stand out in both.

Follow-up:

Compare MSigDB DNA repair, oxidative stress, and EMT signature scores between current smokers and never-smokers in TCGA-LUAD, as violin plots with p-values.

Check the signal at protein level

What to do: Confirm that a transcript-level signal holds at protein level, using proteomics data from CPTAC.

Example prompt:

In CPTAC breast cancer, compare ERBB2 protein abundance between tumors with high and low ERBB2 RNA expression, and show how well RNA and protein levels agree.

Expected result: A protein-level comparison between the two groups, and an RNA-protein concordance plot.

Validate at single-cell resolution

What to do: In K-Pro, use MOSAIC single-cell data to check which cell types drive a bulk signal, for example whether a target is expressed by malignant cells rather than by the tumor microenvironment.

Example prompt:

In MOSAIC bladder cancer, show NECTIN4 expression by cell type, and run a pseudobulk differential expression analysis of malignant cells versus all other cells.

Expected result: A dot plot of NECTIN4 expression across cell types, and a pseudobulk differential expression result for malignant cells.