What problem does it solve? Choosing among the four MOFA implementations (mofapy2, MOFA2, MOFAcellulaR, mofax) and handling cross-cutting concerns like per-view likelihood selection, R-squared semantics, sign ambiguity, and GPU/SVI configuration is error-prone without a shared contract. This Skill acts as the router for the MOFA family, defining the common input/output contract and decision tree so every child skill behaves consistently. ## Core Features & Use Cases - Model-selection decision tree: Routes by working language (Python vs R) and data grain (cells vs donor pseudo-bulk) to the correct mofa-* implementation skill. - Shared contract: Enforces features-by-samples matrices, one likelihood per view (gaussian/bernoulli/poisson), NaN-based missingness, group declaration before build, and explicit seeds. - Cross-cutting references: In-depth documents on architecture, variance-explained semantics, geometric caveats (no biplot, sign/order ambiguity), interop via the shared HDF5 format, and a troubleshooting log. - Use Case: A bioinformatician with RNA log-CPM, binarized ATAC peaks, and raw counts needs to fit a multi-view factor model; this Skill routes them to mofa-mofapy2 with gaussian/bernoulli/poisson likelihoods and the pre-fit checklist before calling run(). ## Quick Start Ask the agent to help you choose the right MOFA implementation for your multi-omics dataset and walk the pre-fit checklist before training.
