AlloGator

Allosteric residue prediction from a single sequence, using protein language-model attention

AlloGator ranks residues by how strongly a protein language model attends from them to the active site. Residues with high attention to the active site, but no direct contact, tend to be allosteric. Provide a protein sequence and the corresponding active site residues (1-indexed), and AlloGator will return a per-residue allosteric score using language model attention.

Citation: Kannan et al., Single-Sequence, Structure-Free Allosteric Residue Prediction with Protein Language Models, bioRxiv (2024). Code & data

Input

0 / 1022 residues
1-indexed positions of the catalytic / active site. Attention is summed toward these.
Enables a 3D view and score-colored structure downloads.