What problem does it solve? Single-cell RNA-seq factor analysis produces latent factors that are hard to interpret: which factors track which biological covariates, which genes drive each factor, and which technical confounders contaminate the decomposition. This Skill operationalises the sciRED Python package to residualise known confounders, extract varimax-rotated factors, and rank factor-covariate associations with a supervised scoreboard. ## Core Features & Use Cases - Confounder residualisation: Fits per-gene Poisson GLMs against a design matrix of technical covariates (library size, protocol, batch) and passes Pearson residuals to PCA. - Varimax-rotated factors: Produces sparse, interpretable gene loadings with rotated cell scores for marker-gene readout. - FCAT and FIST scoreboards: Trains classifier ensembles (LogReg, DecisionTree, XGBoost) to rank factor-covariate associations and combines bimodality, variance, and specificity metrics into an interpretability clustermap. - Use Case: Given an AnnData object with raw counts and cell-type metadata, run the four-step pipeline to discover that factor F3 tracks the stimulated condition while F7 separates cell lines, then export loadings and scores as CSVs. ## Quick Start Use the scired skill to run the sciRED pipeline on my AnnData object with raw counts, residualising library size and protocol, then rank factors against the cell_line covariate.
