Skip to main content

Multiomics skills

Use multiomics skills to analyze Cytos cell screening data with your agent.

Multiomics skills from Element Biosciences guide your agent through Cytos™ cell screening analysis, from loading a Spatial Data Object (SDO) to comparing conditions. A skill describes the steps. Your agent writes and runs the code.

The skills ship as the multiomics plugin in the Elembio/agent-skills repository on GitHub. To add them to your AI app, see Skill installation.

What you need to run analysis​

Skills guide your agent. The code they describe uses the multiomics Python package (elembio-spatialdata-analysis).

Where you analyzeSkillsPython package
A sandbox (ElemBio Catalyst™)Provided by the Sandbox MCPPre-installed
Your own compute (laptop, HPC, or cloud)Install the multiomics pluginInstall elembio-spatialdata-analysis in the environment your agent uses

On your own compute, install both the skills and the Python package. Skills alone are not enough to run analysis.

The skills are validated against elembio-spatialdata-analysis 0.16.0 or later. For changes in each release, see the multiomics skills changelog on GitHub.

What the skills cover​

The skills cover analysis of a Cytos run, from finding the data to comparing conditions:

StageWhat your agent can do
Find and load data
  • Locate a run on local storage or ElemBio Cloud
  • Load an SDO and list its tables, wells, and targets
    • Experiment designInfer conditions, replicates, and treatments from well labels
    Quality control
  • Assess run, well, and tile quality
  • Filter low-quality cells
    • Preprocessing
  • Normalize count data and scale imaging features
  • Assess batch effects
  • Score cell-cycle phase
    • Assay pipelinesRun end-to-end analysis for 3′ transcriptome, OPS, and imaging data
    Clustering and annotation
  • Cluster cells and find marker genes
  • Annotate cell types
  • Validate cluster robustness
    • Differential expression
  • Run pseudobulk differential expression between conditions
  • Plot and interpret the results
    • Multimodal analysis
  • Align 3′ transcriptome, OPS, and imaging data
  • Run joint embeddings and integration across modalities
    • Visualization
  • Plot UMAP and PCA embeddings
  • Plot well images, segmentation overlays, and individual cells

    • Pair skills with Cells2Stats outputs and, when you have an ElemBio Catalyst subscription, a sandbox.

    How your agent uses a skill​

    1. Your agent discovers available skills from the installed plugin or from the Sandbox MCP.
    2. Your agent reads the skill that matches the task.
    3. Your agent follows the skill to generate and run analysis code in an environment that has the multiomics Python package.

    You do not need to copy skill text into chat when the plugin or the Sandbox MCP provides the skills.

    See also​

    Last updated on