simo-multiomics-integration-agent

---

name: 'simo-multiomics-integration-agent' description: 'AI-powered spatial integration of multi-omics datasets using probabilistic alignment for comprehensive tissue atlas construction and cellular state mapping.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools:

• read_file
• run_shell_command

---

SIMO Multiomics Integration Agent

The SIMO Multiomics Integration Agent performs spatial integration of multi-omics datasets through probabilistic alignment. Unlike previous tools limited to transcriptomics, SIMO integrates spatial transcriptomics with single-cell RNA-seq and expands to chromatin accessibility, DNA methylation, and proteomics data.

When to Use This Skill

• When integrating spatial transcriptomics with single-cell multi-omics data.
• For constructing comprehensive tissue atlases with spatial context.
• To map epigenomic states (ATAC-seq, methylation) onto spatial coordinates.
• When analyzing multi-modal cellular phenotypes in tissue architecture.
• For spatial deconvolution combining multiple modalities.

Core Capabilities

1. Spatial-scRNA Integration: Probabilistically align single-cell RNA-seq to spatial coordinates.

2. Chromatin Accessibility Mapping: Project scATAC-seq profiles onto spatial tissue locations.

3. DNA Methylation Spatial Mapping: Integrate single-cell methylation data with spatial context.

4. Multi-Modal Fusion: Combine transcriptomic, epigenomic, and proteomic layers.

5. Probabilistic Cell-Type Assignment: Assign cell types to spatial spots with uncertainty quantification.

6. Spatial Niche Identification: Discover cellular niches defined by multi-omic signatures.

Supported Modalities

Modality	Input Format	Spatial Reference
scRNA-seq	AnnData, Seurat	Visium, MERFISH, Xenium
scATAC-seq	SnapATAC2, ArchR	Visium, Slide-seq
scMethyl	Bismark, allcools	Any spatial modality
CITE-seq (protein)	AnnData	Spatial proteomics
Multi-ome (RNA+ATAC)	Muon, SnapATAC2	All platforms

Integration Algorithm

Step	Method	Purpose
Feature Selection	HVG + marker genes	Reduce dimensionality
Embedding	Variational autoencoder	Shared latent space
Alignment	Optimal transport	Probabilistic matching
Spatial Mapping	Gaussian processes	Smooth spatial predictions
Uncertainty	Posterior sampling	Confidence intervals

Workflow

1. Input: Spatial transcriptomics (Visium/MERFISH/Xenium), reference single-cell multi-omics.

2. Preprocessing: Normalize, select features, QC both datasets.

3. Embedding: Learn joint latent representation across modalities.

4. Probabilistic Alignment: Compute cell-to-spot assignment probabilities.

5. Spatial Imputation: Transfer modalities to spatial coordinates.

6. Niche Analysis: Identify spatial domains by multi-omic signatures.

7. Output: Integrated spatial multi-omics object, niche assignments, visualizations.

Example Usage

User: "Integrate our scRNA-seq and scATAC-seq data with the spatial transcriptomics to understand chromatin states in different tissue regions."

Agent Action:

python3 Skills/Genomics/SIMO_Multiomics_Integration_Agent/simo_integration.py \
    --spatial_data visium_data.h5ad \
    --scrna_ref scrna_atlas.h5ad \
    --scatac_ref scatac_atlas.h5ad \
    --modalities rna,atac \
    --n_spots_per_cell 5 \
    --uncertainty_quantification true \
    --output integrated_spatial_multiome.h5ad

Output Components

Output	Description	Format
Integrated Object	Multi-modal spatial data	AnnData/Muon
Cell Type Map	Spatial cell type assignments	GeoTIFF, CSV
Chromatin Accessibility Map	Spatial ATAC patterns	BigWig, CSV
Niche Assignments	Spatial domain labels	CSV, Zarr
Uncertainty Maps	Per-spot confidence	GeoTIFF
Gene Activity Scores	ATAC-derived gene activity	AnnData layer

Spatial Platforms Supported

Platform	Resolution	Spots/Cells	Genes
10x Visium	55 μm	~5,000	Whole transcriptome
10x Visium HD	8 μm	~300,000	Whole transcriptome
10x Xenium	Subcellular	>100,000	300-5,000 panel
MERFISH	Subcellular	>1M	100-10,000 panel
Slide-seq	10 μm	~60,000	Whole transcriptome
CosMx	Subcellular	>1M	1,000-6,000 panel

AI/ML Components

Variational Integration:

• Multi-modal VAE for joint embeddings
• Contrastive learning for modality alignment
• Batch correction across datasets

Probabilistic Mapping:

• Optimal transport with entropic regularization
• Gaussian process spatial smoothing
• Bayesian uncertainty estimation

Niche Discovery:

• Multi-view clustering
• Spatial autocorrelation (Moran's I)
• Graph neural networks for niche boundaries

Prerequisites

• Python 3.10+
• Scanpy, Squidpy, Muon
• scvi-tools, SnapATAC2
• POT (Python Optimal Transport)
• PyTorch, GPyTorch

Related Skills

• scGPT_Agent - For foundation model embeddings
• Spatial_Epigenomics_Agent - For spatial epigenomics analysis
• Cell_Cell_Communication - For ligand-receptor analysis
• Nicheformer_Spatial_Agent - For spatial niche modeling

Special Considerations

1. Batch Effects: Pre-align datasets from different protocols
2. Spot Deconvolution: Lower resolution platforms need deconvolution
3. Sparsity: scATAC data requires aggregation strategies
4. Compute: Multi-modal integration is memory-intensive
5. Validation: Verify spatial patterns with known marker distributions

Applications

Application	Use Case
Tumor Microenvironment	Map chromatin states of immune infiltrates
Development	Track lineage chromatin dynamics spatially
Neurodegeneration	Spatial mapping of epigenetic changes
Fibrosis	Understand spatial activation programs

Author

AI Group - Biomedical AI Platform