What problem does it solve? Predicting phenotypes from neuroimaging data requires coordinating fMRI connectivity features, sMRI structural features, and a multi-stage deep learning pipeline, which is error-prone to set up manually. This Skill provides model-level guidance for running FM-APP, a foundation model for phenotype prediction via fMRI-to-sMRI knowledge transfer. ## Core Features & Use Cases - Staged Pipeline Guidance: Walks through Stage 0 (data preparation), Stage 1 (fMRI training), Stage 2 (fMRI-T1w alignment via Sinkhorn-RPM), and Stage 3 (decoder and zero-shot inference) with exact script names and expected artifacts. - Delegation to Preprocessing Skills: Routes fMRI ROI extraction to fmri-skill, sMRI feature extraction to smri-skill, and HCP data orchestration to hcpya-skill, keeping model execution separate from preprocessing. - Use Case: A researcher with HCP Young Adult data wants to predict cognitive phenotypes. The Skill instructs them to prepare ROI connectivity features, run the four stages in order, and verify checkpoints like stage1_fmri_best_*.pth before proceeding. ## Quick Start Ask the assistant to run the FM-APP pipeline on your HCP dataset, starting with fMRI ROI feature preparation and proceeding through Stage 1 training, Stage 2 alignment, and Stage 3 zero-shot phenotype inference.
