Batch Correction & Integration Agent
Omi's specialised co-pilot for batch correction work
I diagnose and correct batch effects across your samples, donors, technologies, or timepoints — and I do it without over-correcting away the real biology. The single most underrated step in single-cell analysis, finally done right and explained as we go.
What I can do for you
I diagnose batch effects with kBET, LISI, silhouette scores, and visual UMAP overlays — telling you whether you actually have a batch problem or whether the apparent grouping is real biology you'd be wrong to remove.
I run Harmony, scVI, scANVI, BBKNN, Scanorama, FastMNN, or Seurat CCA/RPCA — and I'll recommend which one fits your design (small batches vs many samples, balanced vs unbalanced, with or without shared cell types).
I benchmark multiple integration methods on your data using scIB metrics (batch correction vs bio conservation) and pick the winner objectively rather than going with whatever was trendy this year.
I integrate across modalities (RNA + ATAC), species (human + mouse with orthologs), or technologies (10x + Smart-seq2) when you need it, and flag when you're asking too much of integration vs needing real experimental controls.
Examples of what you can ask me
Copy any of these straight into the demo, or adapt them to your data.
- 1"Diagnose batch effects in my 12-sample dataset."
- 2"Run Harmony and scVI, then benchmark which integration is better."
- 3"Integrate my human and mouse macrophages using orthologs."
- 4"Is the difference between samples batch or real biology?"
- 5"Integrate 10x v2, v3, and Smart-seq2 datasets of the same tissue."
- 6"Re-integrate without over-correcting my disease vs control signal."
How I work
I run real Scanpy (Python) or Seurat (R) code on the secure MCP server — no hallucinations, no made-up gene lists. Every result comes with the exact code I executed and the parameters I used, so your analysis is fully reproducible and ready for the Methods section.
Best for
Multi-sample studies, multi-center collaborations, atlas projects, anyone combining old and new datasets, and labs whose UMAPs separate by sample instead of by cell type (we've all been there).
References
- Harmony (Korsunsky et al., 2019) – Nature Methods
- scVI / scANVI (Lopez et al., 2018; Xu et al., 2021) – Nature Methods
- BBKNN (Polański et al., 2020) – Bioinformatics
- Scanorama (Hie et al., 2019) – Nature Biotechnology
- FastMNN (Haghverdi et al., 2018) – Nature Biotechnology
- scIB benchmark (Luecken et al., 2022) – Nature Methods
- kBET (Büttner et al., 2019) – Nature Methods
Try Batch Correction now
Jump into the demo with a starter prompt already loaded. Upload your data, or play with our example dataset first.