Available tools
Capability map of what PhenoMe can do. For method names and signatures, see the Function location guide and the API reference.
Most analysis methods accept a feature source: "embeddings", "properties", or "combined".
| Category | Tools / options | Notes |
|---|---|---|
| Image I/O | TIFF, NIfTI, NumPy (.npy/.npz), PNG/JPEG/BMP |
tifffile, nibabel, numpy, OpenCV; discovery also matches many microscopy extensions (e.g. .czi, .nd2) without dedicated readers |
| Embeddings | DINOv2 (built-in); custom via ModelWrapper |
e.g. dinov2_vitb14_reg; channel modes "split" / "combined" |
| Dimensionality reduction | PCA, t-SNE, UMAP | CPU: sklearn + umap-learn; optional GPU: TorchDR (use_gpu_for_dr=True) |
| Clustering | k-means, GMM, DBSCAN | Optional normalize, silhouette, and PCA/t-SNE/UMAP pre-reduce |
| Distances & similarity | Euclidean, cosine; prototypes; outliers | Reference modes centroid / all_to_all; outliers via z-score or IQR |
| Group enrichment | Leave-one-out z-scores; Welch + BH-FDR; property stats | Also top properties vs reference (cohens_d / mean_diff) |
| Properties | Presets: none, intensity, shape, basic, standard, complete | Backends: scikit-image, scipy, numpy; factories for regionprops, intensity, texture, rings, blur, entropy; blob plugin |
| Batch correction | Sphering, z-score | Control-based per batch/plate (correct_batches) |
| Interpretability | Pearson, Spearman, dCor, MI; lasso, random forest | Correlate or explain DR/embedding axes with properties |
| Visualization & reporting | Plotly plots; matplotlib image grids; Jupyter explorer; HTML report | DR scatters, enrichment, correlations, gallery, etc. |
| Preprocessing & I/O | Transforms; L2 / StandardScaler; HDF5 checkpoints; CSV/Parquet/Excel export | Orchestrated on PhenoMe |
See also: Embeddings, Distances, Select properties, Visualization & results, Embeddings interpretability.