Separating benchmark discrimination from experimental retrieval in synthetic-lethality prediction

In five gene-disjoint folds, degree-matched benchmark negatives drove AUROC for SLMGAE from 0.782 to 0.501, showing benchmark discrimination can shift strongly with test composition.

• Why it matters: Degree-aware evaluation can reveal sampling bias, but it is unclear how a degree-controlled benchmark score should be interpreted for experimental screening priorities.
• What they did: The authors linked fixed-prediction benchmark interventions to retrieval in measured experimental pools, using crossed training-testing plus a three-seed diagnostic across five gene-disjoint folds.
• The result: In four measured screens, Random-trained SLMGAE retrieved 21 SPIDR hits in 198 selections versus 8.00 expected (2.62-fold enrichment; Holm-adjusted p=0.0008), indicating near-chance matched-test AUROC can still support useful experimental prioritization.

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