scfoundation-model-agent

name: 'scfoundation-model-agent' description: 'Unified agent for leveraging single-cell foundation models (scGPT, scBERT, Geneformer, scFoundation) for cross-species annotation, perturbation prediction, and gene network inference.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools: - read_file - run_shell_command

scFoundation Model Agent

The scFoundation Model Agent provides a unified interface to leverage state-of-the-art single-cell foundation models for diverse downstream tasks. It integrates scGPT, scBERT, Geneformer, scFoundation, and emerging models to enable cross-species cell annotation, in silico perturbation prediction, gene regulatory network inference, and batch integration.

When to Use This Skill

• When annotating cell types across species (human, mouse, cross-species).
• For predicting perturbation effects (knockouts, drug treatments) in silico.
• To infer gene regulatory networks from single-cell data.
• When integrating batches without losing biological signal.
• For generating cell embeddings for downstream analysis.

Core Capabilities

1. Cross-Species Cell Annotation: Transfer cell type labels across species using unified embeddings.

2. In Silico Perturbation: Predict gene expression changes from knockouts/treatments.

3. Gene Regulatory Network Inference: Discover TF-target relationships from attention patterns.

4. Batch Integration: Remove technical variation while preserving biology.

5. Cell Embedding Generation: Generate universal cell representations for any downstream task.

6. Multi-Model Ensemble: Combine predictions from multiple foundation models.

7. LLM-Based BGC Discovery (Reference): DeepSeMS (Xu et al., Nat Comput Sci, Apr 2026) demonstrates a large language model approach for revealing hidden biosynthetic potential of the global ocean microbiome. Complementary reference for applying LLM/foundation-model methods to biosynthetic gene cluster (BGC) discovery from microbiome metagenomic data, extending foundation-model coverage beyond single-cell transcriptomics. PMID 42062603.

8. scPRINT-2 (Next-Gen Single-Cell Foundation Model): scPRINT-2 (cantinilab, GitHub repo, Apr 2026) is a next-generation single-cell foundation model alongside scGPT/scBERT/Geneformer/scFoundation. Intended for large-scale single-cell perturbation prediction, cell annotation, and gene regulatory network inference. Distributed as a Jupyter Notebook reference implementation; consult the upstream repository for licensing terms before use. https://github.com/cantinilab/scPRINT-2

9. Stack (In-Context Learning Single-Cell Foundation Model): Stack (Arc Institute, GitHub repo, Apr 2026) is a single-cell foundation model that supports in-context learning at inference time, removing the need for task-specific fine-tuning when adapting to new datasets. Important pattern shift versus prior fine-tuning-based foundation models — enables few-shot cell type annotation and perturbation prediction across previously unseen datasets by conditioning on context examples at inference. https://github.com/ArcInstitute/stack

10. Phenotype-Associated Subpopulation Identification: Zhao et al. (BMC Biol, Apr 2026, PMID 42050525) describe a workflow combining transfer learning from single-cell foundation models with statistical ensemble learning to identify phenotype-associated cell subpopulations. Use foundation-model embeddings as the transfer substrate when phenotype signal may reflect cell states within or across standard clusters; validate candidates through ensemble stability/resampling and explicit phenotype-association checks rather than treating Leiden/UMAP clusters alone as phenotype-linked discoveries.

Supported Foundation Models

Workflow

1. Input: Single-cell RNA-seq data (AnnData format).

2. Model Selection: Choose appropriate model(s) for task.

3. Preprocessing: Tokenize genes, normalize expression.

4. Inference: Generate embeddings or predictions.

5. Task Execution: Annotation, perturbation, or network inference.

6. Ensemble (Optional): Combine multi-model predictions.

7. Output: Annotated data, predictions, networks.

Example Usage

User: "Use scGPT to predict the effect of CRISPR knockout of TP53 on these cancer cells."

Agent Action:

python3 Skills/Genomics/scFoundation_Model_Agent/foundation_predict.py \
    --input cancer_cells.h5ad \
    --model scgpt \
    --task perturbation \
    --perturbation "TP53 knockout" \
    --model_checkpoint scgpt_human_gene_v1.pt \
    --output tp53_ko_predictions.h5ad

Task-Specific Usage

Cell Type Annotation

python3 foundation_predict.py \
    --input query_cells.h5ad \
    --model geneformer \
    --task annotation \
    --reference tabula_sapiens.h5ad \
    --output annotated_cells.h5ad

Gene Network Inference

python3 foundation_predict.py \
    --input cells.h5ad \
    --model scgpt \
    --task grn_inference \
    --transcription_factors tf_list.txt \
    --output gene_network.csv

Batch Integration

python3 foundation_predict.py \
    --input multi_batch.h5ad \
    --model scfoundation \
    --task integration \
    --batch_key batch \
    --output integrated.h5ad

Phenotype-Associated Subpopulation Identification

Transfer-learn a single-cell foundation model on labeled reference cells, project phenotype labels (disease status, treatment response, etc.) onto query embeddings, then run statistical ensemble learning over multiple resamples/models to flag robust phenotype-linked subpopulations. Use this mode when the goal is phenotype association rather than only cell-type annotation: confirm that selected subpopulations remain stable across ensemble runs and show phenotype association beyond standard clustering boundaries. Pattern follows Zhao et al. (BMC Biol, 2026, PMID 42050525).

python3 foundation_predict.py \
    --input query_cells.h5ad \
    --model scfoundation \
    --task phenotype_subpopulation \
    --reference labeled_reference.h5ad \
    --phenotype_key disease_status \
    --transfer_mode finetune \
    --ensemble_runs 25 \
    --ensemble_method bootstrap_vote \
    --output phenotype_subpopulations.h5ad

Output Formats

Performance Benchmarks

AI/ML Architecture

Transformer Backbone:

• Gene-level tokenization
• Attention-based gene interactions
• Masked expression prediction pretraining

Perturbation Module:

• Conditional generation
• Counterfactual prediction
• Dose-response modeling

Transfer Learning:

• Zero-shot annotation
• Few-shot fine-tuning
• Domain adaptation

Prerequisites

• Python 3.10+
• PyTorch 2.0+
• transformers, flash-attn
• Scanpy, AnnData
• Model-specific weights
• GPU with 16GB+ VRAM

Related Skills

• Nicheformer_Spatial_Agent - For spatial foundation models
• scGPT_Agent - Dedicated scGPT workflows
• Cell_Type_Annotation - Traditional annotation methods
• Pathway_Analysis - Gene set enrichment

Model Selection Guide

Special Considerations

1. Gene Coverage: Models trained on variable gene sets; check overlap
2. Species: Some models human-only; use UCE for cross-species
3. Compute: Large models need significant GPU memory
4. Fine-Tuning: Task-specific fine-tuning improves performance
5. Versioning: Model weights update frequently; track versions

Ensemble Strategies

Author

AI Group - Biomedical AI Platform

References

• Xu T, Yang Y, Zhu R, Lin W, Li J. DeepSeMS: revealing the hidden biosynthetic potential of the global ocean microbiome with a large language model. Nat Comput Sci, 2026 Apr 30. https://pubmed.ncbi.nlm.nih.gov/42062603/
• cantinilab. scPRINT-2: next-gen single-cell foundation model. GitHub, 2026 Apr 29. https://github.com/cantinilab/scPRINT-2
• Arc Institute. Stack: single-cell foundation model with in-context learning at inference time. GitHub, 2026 Apr 28. https://github.com/ArcInstitute/stack
• Zhao Y, Pan X, Luo Z, Liu Q. Single-cell phenotype-associated subpopulation identification via transfer foundation model and statistical ensemble learning. BMC Biol, 2026 Apr 29. https://pubmed.ncbi.nlm.nih.gov/42050525/