Benchmark results

Leaderboard

Rank models across the included datasets using each dataset's default split. Choose an overall, prediction-task, or dataset view.

Complete overall rankings
--
Models evaluated
--
Datasets
--
Model-target summaries
--

Loading coverage summary.

Loading leaderboard

How ranking works

From split seeds to mean z-score

  1. 01

    Standardize each target

    Within each seed, dataset, and target, model performance is converted to a z-score. Pearson correlations receive a Fisher transform first.

  2. 02

    Combine related results

    Within each seed, z-scores are averaged across targets, then across related sub-datasets such as cell lines or species.

  3. 03

    Average split seeds

    Each model is evaluated on ten train-test splits. The aggregated z-scores are then averaged across those seeds.

  4. 04

    Compute the overall score

    Complete models are ranked by their mean z-score across the 11 source datasets represented by the 19 registered dataset IDs.

Standardization makes AUPRC and Pearson correlation comparable only as relative performance within the same target. The overall score is the manuscript's mean standardized metric, not an absolute measure of biological performance. Regression uses cross-validated Ridge results.