Gene-level structure of prior knowledge and perturbation responses informs target-level predictability in prior-informed models

Prior agreement among eight sources of biological knowledge explains 30–47% of variance in co-function agreement and 35–43% in co-regulation agreement across genome-wide and essential-gene Perturb-seq screens.

• Why it matters: It is unclear whether prior knowledge captures target-specific perturbation-response relationships and whether prior-derived metrics can forecast target-level predictive performance without measuring responses.
• What they did: The study builds a gene-similarity framework comparing agreement across eight prior knowledge sources and co-function/co-regulation signals from six Perturb-seq datasets, then links prior agreement and prior-derived train-test similarity to target-level prediction accuracy in three prior-informed models.
• The result: Predictions are more accurate for targets whose observed perturbation-response relationships better match prior knowledge and are more similar to training perturbations, and prior agreement and test-free train-test similarity track relative target-level predictive performance differences.

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