polyprotein_stats

A small tool to test and visualize protein embeddings and amino acid proportions.

Currently deployed on streamlit.io.

Given a set of proteins, this tool seeks to answer these questions:

• is there a enrichment of one or two amino acids?
• can amino acid proportions, plus sequence length discriminate the input proteins from the rest of the proteome?
• can the averaged per residue embeddings from a deep neural network discriminate the input proteins from the rest of the proteome?

Embeddings for the human proteins is from the ProtT5 embedder at full precision (prottrans_t5_xl_u50 model) that were generated by Christian Dallago and Burkhard Rost (Zenodo). Also, see 'ProtTrans: Towards Cracking the Language of Life’s Code Through Self-Supervised Learning' by Elnaggar et al. biorxiv preprint