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
Spatial-scRNA Integration: Probabilistically align single-cell RNA-seq to spatial coordinates.
Chromatin Accessibility Mapping: Project scATAC-seq profiles onto spatial tissue locations.
DNA Methylation Spatial Mapping: Integrate single-cell methylation data with spatial context.
Multi-Modal Fusion: Combine transcriptomic, epigenomic, and proteomic layers.
Probabilistic Cell-Type Assignment: Assign cell types to spatial spots with uncertainty quantification.
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
Input: Spatial transcriptomics (Visium/MERFISH/Xenium), reference single-cell multi-omics.
Preprocessing: Normalize, select features, QC both datasets.
Embedding: Learn joint latent representation across modalities.
Probabilistic Alignment: Compute cell-to-spot assignment probabilities.
Spatial Imputation: Transfer modalities to spatial coordinates.
Niche Analysis: Identify spatial domains by multi-omic signatures.
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
- Batch Effects: Pre-align datasets from different protocols
- Spot Deconvolution: Lower resolution platforms need deconvolution
- Sparsity: scATAC data requires aggregation strategies
- Compute: Multi-modal integration is memory-intensive
- 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
1---2name: simo-multiomics-integration-agent3description: AI-powered spatial integration of multi-omics datasets using probabilistic alignment for comprehensive tissue atlas construction and cellular state mapping.4license: MIT5---67# SIMO Multiomics Integration Agent89The **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.1011## When to Use This Skill1213* When integrating spatial transcriptomics with single-cell multi-omics data.14* For constructing comprehensive tissue atlases with spatial context.15* To map epigenomic states (ATAC-seq, methylation) onto spatial coordinates.16* When analyzing multi-modal cellular phenotypes in tissue architecture.17* For spatial deconvolution combining multiple modalities.1819## Core Capabilities20211. **Spatial-scRNA Integration**: Probabilistically align single-cell RNA-seq to spatial coordinates.22232. **Chromatin Accessibility Mapping**: Project scATAC-seq profiles onto spatial tissue locations.24253. **DNA Methylation Spatial Mapping**: Integrate single-cell methylation data with spatial context.26274. **Multi-Modal Fusion**: Combine transcriptomic, epigenomic, and proteomic layers.28295. **Probabilistic Cell-Type Assignment**: Assign cell types to spatial spots with uncertainty quantification.30316. **Spatial Niche Identification**: Discover cellular niches defined by multi-omic signatures.3233## Supported Modalities3435| Modality | Input Format | Spatial Reference |36|----------|--------------|-------------------|37| scRNA-seq | AnnData, Seurat | Visium, MERFISH, Xenium |38| scATAC-seq | SnapATAC2, ArchR | Visium, Slide-seq |39| scMethyl | Bismark, allcools | Any spatial modality |40| CITE-seq (protein) | AnnData | Spatial proteomics |41| Multi-ome (RNA+ATAC) | Muon, SnapATAC2 | All platforms |4243## Integration Algorithm4445| Step | Method | Purpose |46|------|--------|---------|47| Feature Selection | HVG + marker genes | Reduce dimensionality |48| Embedding | Variational autoencoder | Shared latent space |49| Alignment | Optimal transport | Probabilistic matching |50| Spatial Mapping | Gaussian processes | Smooth spatial predictions |51| Uncertainty | Posterior sampling | Confidence intervals |5253## Workflow54551. **Input**: Spatial transcriptomics (Visium/MERFISH/Xenium), reference single-cell multi-omics.56572. **Preprocessing**: Normalize, select features, QC both datasets.58593. **Embedding**: Learn joint latent representation across modalities.60614. **Probabilistic Alignment**: Compute cell-to-spot assignment probabilities.62635. **Spatial Imputation**: Transfer modalities to spatial coordinates.64656. **Niche Analysis**: Identify spatial domains by multi-omic signatures.66677. **Output**: Integrated spatial multi-omics object, niche assignments, visualizations.6869## Example Usage7071**User**: "Integrate our scRNA-seq and scATAC-seq data with the spatial transcriptomics to understand chromatin states in different tissue regions."7273**Agent Action**:74```bash75python3 Skills/Genomics/SIMO_Multiomics_Integration_Agent/simo_integration.py \76 --spatial_data visium_data.h5ad \77 --scrna_ref scrna_atlas.h5ad \78 --scatac_ref scatac_atlas.h5ad \79 --modalities rna,atac \80 --n_spots_per_cell 5 \81 --uncertainty_quantification true \82 --output integrated_spatial_multiome.h5ad83```8485## Output Components8687| Output | Description | Format |88|--------|-------------|--------|89| Integrated Object | Multi-modal spatial data | AnnData/Muon |90| Cell Type Map | Spatial cell type assignments | GeoTIFF, CSV |91| Chromatin Accessibility Map | Spatial ATAC patterns | BigWig, CSV |92| Niche Assignments | Spatial domain labels | CSV, Zarr |93| Uncertainty Maps | Per-spot confidence | GeoTIFF |94| Gene Activity Scores | ATAC-derived gene activity | AnnData layer |9596## Spatial Platforms Supported9798| Platform | Resolution | Spots/Cells | Genes |99|----------|------------|-------------|-------|100| 10x Visium | 55 μm | ~5,000 | Whole transcriptome |101| 10x Visium HD | 8 μm | ~300,000 | Whole transcriptome |102| 10x Xenium | Subcellular | >100,000 | 300-5,000 panel |103| MERFISH | Subcellular | >1M | 100-10,000 panel |104| Slide-seq | 10 μm | ~60,000 | Whole transcriptome |105| CosMx | Subcellular | >1M | 1,000-6,000 panel |106107## AI/ML Components108109**Variational Integration**:110- Multi-modal VAE for joint embeddings111- Contrastive learning for modality alignment112- Batch correction across datasets113114**Probabilistic Mapping**:115- Optimal transport with entropic regularization116- Gaussian process spatial smoothing117- Bayesian uncertainty estimation118119**Niche Discovery**:120- Multi-view clustering121- Spatial autocorrelation (Moran's I)122- Graph neural networks for niche boundaries123124## Prerequisites125126* Python 3.10+127* Scanpy, Squidpy, Muon128* scvi-tools, SnapATAC2129* POT (Python Optimal Transport)130* PyTorch, GPyTorch131132## Related Skills133134* scGPT_Agent - For foundation model embeddings135* Spatial_Epigenomics_Agent - For spatial epigenomics analysis136* Cell_Cell_Communication - For ligand-receptor analysis137* Nicheformer_Spatial_Agent - For spatial niche modeling138139## Special Considerations1401411. **Batch Effects**: Pre-align datasets from different protocols1422. **Spot Deconvolution**: Lower resolution platforms need deconvolution1433. **Sparsity**: scATAC data requires aggregation strategies1444. **Compute**: Multi-modal integration is memory-intensive1455. **Validation**: Verify spatial patterns with known marker distributions146147## Applications148149| Application | Use Case |150|-------------|----------|151| Tumor Microenvironment | Map chromatin states of immune infiltrates |152| Development | Track lineage chromatin dynamics spatially |153| Neurodegeneration | Spatial mapping of epigenetic changes |154| Fibrosis | Understand spatial activation programs |155156## Author157158AI Group - Biomedical AI Platform
diegosouzapw