Use ESM-IF1

Overview

The ESM-IF1 inverse folding model predicts protein sequences from their backbone atom coordinates, trained with 12M protein structures predicted by AlphaFold2.

Features

  • Allows you to inverse fold any protein or complex of proteins.
  • Includes options to control which chains to design and which to keep fixed.
  • Supports different sampling techniques to better explore the protein landscape.
  • Includes per sequence metrics such as an overall score and sequence recovery.
  • Includes amino acid probabilities by position.
View Paper

Configuration & Options

Model Inputs

The input protein structure to predict the amino acid sequence of. Acceptable input file formats include pdb and mmcif.

Specify the name of the chain(s) that you want to inverse fold and predict new sequences for. The provided name needs to match the name on the pdb/cif file. For multiple chains seperate them with a comma (e.g., "A,B,C")

The number of output sequences to generate.

Advanced Options

Lower sampling temperature typically results in higher sequence recovery but less diversity. If you set your temperature too low then all your samples will be the same. Ideally we recommend something in between 0.1-0.5.

Check this option to enable "Training Mode" which activates drop out layers within the model. This can be used to force the model to create more diverse predictions, but at the potential cost of accuracy.

Ready to submit your job?

Once you're done just hit the submit button below and let us do the rest.

To submit a job please login or register an account.