Transcriptomic similarity guides source selection beyond tissue lineage

Baseline transcriptomic similarity improves source selection for held-out targets beyond tissue lineage, yielding drug-balanced ranking effects of 0.359 (95% CI 0.337–0.383) in GDSC2.

• Why it matters: Perturbational and pharmacogenomic maps are incomplete across cellular contexts, so it is unclear which measured context should guide response prediction for an unmeasured target.
• What they did: Candidate sources were ranked by baseline transcriptomic similarity, then evaluated separately from aggregation and transfer-uncertainty choices using held-out targets in LINCS, GSE306429, MIX-Seq, and a full GDSC2 analysis.
• The result: Similarity-linked ranking utility exceeded random in GSE306429 (ΔU = 0.034; P = 0.002) and MIX-Seq exact-time tasks (ΔU = 0.046), and remained positive under lineage restriction (0.130; 95% CI 0.119–0.141).

AI summary of the abstract · may contain errors · not medical advice · read the paper.