Our new preprint is out: biorxiv.org/content/10.6...
We show that GRN benchmarks commonly used overestimates performance due to negative-sampling choices, sometimes allowing simple degree-based baselines to rival more complex GNNs.
A reminder of why bias-aware evaluation matters!
We show that GRN benchmarks commonly used overestimates performance due to negative-sampling choices, sometimes allowing simple degree-based baselines to rival more complex GNNs.
A reminder of why bias-aware evaluation matters!