Skip to content

README.md

Latest commit

 

History

History
33 lines (22 loc) · 1.78 KB

README.md

File metadata and controls

33 lines (22 loc) · 1.78 KB

Protein Language Model (PLM) Embeddings Generation

This directory contains SLURM-compatible files to generate protein language model embeddings from various models. These files run evolvepro/plm/[protein_language_model]/extract.py to generate embeddings in CSV format from a FASTA file containing all single amino acid mutants of interest.

How to Run

The usage of extract.py files may vary depending on the model. Below is a general example of how to use the scripts:

python evolvepro/plm/[model_name]/extract.py [model_location] [fasta_file] [output_dir] [additional_options]

Example:

To generate a CSV file of ESM-2 (15B parameter model) embeddings of the last hidden layer from a FASTA file of single amino acid mutants:

python evolvepro/plm/esm/extract.py esm2_t48_15B_UR50D output/dms/brenan.fasta output/plm/esm/brenan --toks_per_batch 512 --include mean --concatenate_dir output/plm/esm/

In this example:

  • esm2_t48_15B_UR50D is the model name
  • output/dms/brenan.fasta is the input FASTA file
  • output/plm/esm/brenan is the output directory for the embeddings
  • --toks_per_batch 512 sets the number of tokens per batch
  • --include mean specifies to include mean pooled embeddings
  • --concatenate_dir output/plm/esm/ sets the directory for concatenated results

Note: Adjust --toks_per_batch based on your GPU memory capacity.

Model-Specific Instructions

Different protein language models may require specific commands or have unique parameters. Refer to the extract.py files in each model's subdirectory and/or the SLURM-compatible files in this directory for model-specific instructions. We recommend using a GPU powered HPC to extract protein language model embeddings, but these SLURM-compatible files can be easily adapted into a command line call.