Core concepts

Use this section when you want background on how PhenoMe represents phenotyping: embeddings, channel handling, distances, metadata, and what gets stored in results. You can install, run the Getting started workflow, and read the Learning paths without it; come back here when terminology in other pages needs a fuller explanation.

What is phenotyping?

Phenotyping characterizes the observable traits of biological samples. In microscopy this means analyzing cell shape (morphology), where proteins are located (fluorescence patterns), or how structures change under treatment.

PhenoMe automates this with deep learning. It is model-agnostic (it works with almost any vision model) and dataset-agnostic (it does not care how your files are organized).

Processing flow

PhenoMe processes data through a bifurcated 4-step pipeline:

1. Discovery (Input) — Discover images and extract experimental metadata.
2. Embeddings (A) — Load images, apply transforms, and extract deep learning embeddings.
3. Properties (B) — Compute interpretable classical features (shape, texture, intensity) in parallel.
4. Analysis & Visualization (C) — Integrate both sources to quantify phenotypic distances and produce interactive plots.

Explore concepts

Embeddings & ModelsHow visual fingerprints are extracted and the lifecycle of embeddings.

Channel modesHow to handle multi-channel fluorescence vs. brightfield images.

Distance metricsMeasuring phenotypic change between treated and control groups.

Metadata & propertiesLinking images to experimental context and classical image features.

Property interpretationWhat each built-in classical property measures (shape, intensity, texture, QC) and how to avoid redundant comparisons.

Visualization & resultsDimensionality reduction (PCA/UMAP), results container structure, and sharing or portability of checkpoints.

Embeddings interpretabilityCorrelating embeddings and DR axes with classical properties: univariate and multivariate explainability.