Scfoundation Model Agent

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.

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

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