miRBench Leaderboard

This leaderboard compares models evaluated on the miRBench test sets.

Submission rules

Every submission must include:

• Publicly accessible source code with instructions for reproducing the results.
• Downloadable model weights.
• Prediction files for both the test set and the left-out test set.
• Test AUPRC and left-out AUPRC values calculated from the submitted predictions.
• The model name and author or team name.

Leaderboard miRBench v7

Model name	Author	Code	Weights	Predictions	Test AUPRC	Left-out AUPRC
PHACT-P1 Multi-Candidate Fusion Agentomics	Dimos	Code	Weights	Predictions	0.890	0.861
PHACT-P1 Multimodal Fusion Agentomics	Dimos	Code	Weights	Predictions	0.890	0.862
PHACT-P1 miRNA + Target + Conservation CNN	Dimos	Code	Weights	Predictions	0.887	0.866
PHACT-P1 miRNA + Target CNN	Dimos	Code	Weights	Predictions	0.886	0.865
PHACT-P1 miRNA CNN	Dimos	Code	Weights	Predictions	0.878	0.870
miRBind2 Retrained v7	David/Panos	Code	Weights	Predictions	0.878	0.870
* miRBind2 V5	Paper	Code			0.85	0.83
* DeepExoMir	Paper	Code			0.848
* CNN retrained (Manakov2022)	Paper	Code			0.84	0.81
* MiRformer, Manakov2022-trained	Paper	Code			0.83	0.70

• not tested on v7

The five PHACT-P1 entries use the consensus parameter-1 mature-miRNA profiles and supersede the earlier experimental PHACT/miRBind entries.