scMultiBench

A systematic benchmark of single-cell multimodal integration. Run 40+ integration methods through one typed Python API, score them with scIB and task-specific metrics, and explore the rankings to pick the right method for your data.

pip install multibench-sc

Get started How it works

1 Run a method

Run Run a method Drive any of 40+ integration methods through one typed API, each in its own environment.

Evaluate Evaluate a run Score a run output with scIB batch-correction, biological-conservation, and task-specific metrics.

Plot Plot results Load published results into tidy frames and render scIB-style bubble plots and rankings.

Explore Interactive explorer Browse benchmark results, metrics, rankings, and bubble tables in the Shiny explorer.

40+ methods, one API Run integration methods across four scenarios, vertical, diagonal, mosaic, and cross, through the typed multibench API. Each method runs in its own conda env, so their conflicting toolchains never collide.

Metrics you can trust Each task gets its own metric panel, scIB batch-correction and biological-conservation scores, classification accuracy and F1, marker overlap, RMSE, plus runtime and peak memory, aggregated into per-method grand rank scores.

Choose with evidence No single method wins everywhere. Explore rankings, bubble tables, and a decision tree that recommends top methods per scenario and task, so method choice is grounded in your data, not folklore.