foundation-model-analysis

Single-cell perturbation prediction uses foundation models to forecast cellular responses. This skill profiles data, selects a compatible model, and executes predictions for large-scale perturbation scenarios across diverse datasets.

Comparing foundation models for single-cell integration requires profiling data and checking compatibility across 22 supported models like scGPT and UCE. This skill enforces a 6-step pipeline with cross-model comparisons and provenance tracking for reproducible results.

Cross-species datasets are supported for single-cell foundation model workflows. The pipeline applies embeddings, annotations, and integration tasks to diverse datasets including human and model species through a unified API.

Checkpoint management for foundation models on single-cell data is handled automatically with per-model tracking. The workflow enforces provenance tracking and device selection to ensure reproducible embeddings and annotations.

Drug response prediction is supported as a task within single-cell foundation model workflows. The pipeline coordinates model selection and execution for drug_response tasks alongside embedding, annotation, integration, and perturbation predictions.

A compatibility check before running single-cell embeddings ensures the selected foundation model aligns with your dataset profile. This validation step prevents execution failures and guarantees reproducible results across different models.