What am I looking at?
Each point is one protein: the identity vector the model learned for it during pretraining, reduced to two dimensions. Proteins the model treats alike sit close together.
The model never saw a sequence, a family or a pathway. It only saw which proteins tend to be quantified together and rise and fall together across 45,188 proteomes. That alone is enough for ribosomal proteins, histones, keratins, immunoglobulins and other families to gather into islands. Proteins that are rarely detected have had little chance to learn anything and stay in the diffuse centre; slide the detection threshold up to watch the structure sharpen.
There are 39 annotated families here. Many more islands are visible than we have named: a tight group of grey points is a cluster the model found that nobody has labelled yet. Look up your protein of interest and check its neighbours; they may be partners no database lists.
Hovera point to read gene, family and detection rate.
Clickto pin it and see its twenty nearest proteins in the model’s space.
Familyswitch colouring to family and pick one to see it in lime against the rest.
Searcha gene (RPL3, KRT5) or a KEGG pathway (Ribosome, Oxidative phosphorylation) to light it up.
