polishing

Author: timtreis

content/blog/2025-07-biomni.md

@@ -1,5 +1,5 @@
 +++
-title = "scverse × Biomni: Agentic Single-Cell Omics Analysis"
+title = "scverse × Biomni: Agentic Single-Cell Analysis"
 date = 2025-07-28T00:00:05+01:00
 description = "Biomni, the general-purpose biomedical AI agent now supports scverse tooling."
 author = "scverse team & Biomni team"
@@ -12,7 +12,9 @@ Single-cell and spatial omics have unlocked unprecedented insights into cellular
 
 **scverse** is a community-driven, open-source initiative behind many of the most widely adopted Python tools in single-cell biology, known for promoting modular, interoperable, and scalable analysis across diverse modalities—from transcriptomics to spatial and immune profiling.
 
-We’re excited to announce a collaboration between **scverse** and **Biomni** to further streamline and enhance single-cell and spatial omics analyses. Biomnis intelligent agentic interface os now capable of using scverse tools, enabling researchers to seamlessly integrate, execute, and manage analyses across ten core scverse packages through natural language prompts. Researchers can describe their analysis goals in plain English—e.g., *“cluster cells and identify markers”* or *“analyze perturbation effects between treatment groups”*—and Biomni automatically generates and runs the corresponding scanpy, pertpy, squidpy, or scvi-tools code. The agent understands biological context, handles parameters and dependencies, and returns reproducible results—no coding required. Notably, all codes written by agents are documented and available to users for reproduction or modification as a Jupyter notebook.  This is an early-stage capability, and manual review of outputs is encouraged to ensure accuracy, and we invite constructive community feedback to improve. 
+We’re excited to announce a collaboration between **scverse** and **Biomni** to further streamline and enhance single-cell and spatial omics analyses. Biomnis intelligent agentic interface is now capable of using scverse tools, enabling researchers to seamlessly integrate, execute, and manage analyses across ten core scverse packages through natural language prompts. Researchers can describe their analysis goals in plain English - e.g., *“cluster cells and identify markers”* or *“analyze perturbation effects between treatment groups”* - and Biomni automatically generates and runs the corresponding scanpy, pertpy, squidpy, or scvi-tools code. 
+
+The agent understands biological context, handles parameters and dependencies, and returns reproducible results - no coding required. Notably, all codes written by agents are documented and available to users for reproduction or modification as a Jupyter notebook.  This is an early-stage capability, and manual review of outputs is encouraged to ensure accuracy, and we invite constructive community feedback to improve. 
 
 ---
 
@@ -24,9 +26,9 @@ To illustrate, here are a few examples of how Biomni equipped with scverse packa
 
    → Biomni invokes [Scanpy](https://scanpy.readthedocs.io/en/latest/) for preprocessing, clustering, annotation, and visualization.
 
-* *“Perform compositional analysis with scCODA in this perturb-seq dataset.”*
+* *“Perform compositional analysis with [scCODA](https://sccoda.readthedocs.io/en/latest/) in this perturb-seq dataset.”*
 
-   → Biomni uses [Pertpy](https://github.com/scverse/pertpy/) to perform scCODA compositional analysis.
+   → Biomni uses [Pertpy](https://github.com/scverse/pertpy/) to perform [scCODA](https://sccoda.readthedocs.io/en/latest/) compositional analysis.
 
 * *“Detect spatially variable genes in this Visium dataset and overlay results on the H\&E image.”*
 
@@ -55,7 +57,7 @@ Moving beyond the agentic delegation of individual single-cell tasks, Biomni als
 - [✓] Driver pathway identification (completed)
 - [✓] Generate comprehensive HTML report (completed)
 
-<img src="/img/blog/scverse_x_biomni_ui.png" style="max-width: 100%;" alt="scverse × Biomni partnership banner" />
+<img src="/img/blog/scverse_x_biomni_ui.png" style="max-width: 100%;" alt="Biomni interface" />
 
 Biomni has successfully completed these steps in around 20 minutes. It preprocesses the data, identifies 21 distinct cell types from 17,771 high-quality cells, and maps them across spatial axes with clear anterior-posterior and dorsal-ventral organization. It uncovers major tissue compartments, quantifies gene expression variability (e.g., Slc4a1, T, Mesp2), and performs pathway enrichment, highlighting nervous system and cation channel activity. Biomni also infers cell-cell interactions, such as cardiomyocyte clustering and endothelial co-localization with hematopoietic progenitors. This entire sequence is automatically orchestrated by the agent, enabling reproducible and modular single-cell workflows without writing a single line of code.
 
@@ -113,4 +115,5 @@ By combining data-driven discovery with biological reasoning, Biomni helps resea
 
 ---
 
+### Disclaimer
 This integration between Biomni and scverse highlights a powerful synergy, removing coding barriers while maintaining transparency, reproducibility, and scientific rigor. Although agent-based automation enhances convenience, human expertise remains vital for biological interpretation and validation. We look forward to seeing the innovative discoveries our community achieves with this exciting collaboration.