Skip to content

Latest commit

 

History

History
59 lines (45 loc) · 2.63 KB

README.md

File metadata and controls

59 lines (45 loc) · 2.63 KB
Raw

Deep Mutational Scanning (DMS) Workflow for EVOLVEpro

This directory contains scripts for running the DMS workflow of EVOLVEpro, which optimizes a few-shot model on deep mutational scanning datasets. The source functions used here are in evolvepro/src/evolve.py, which calls the underlying model, in evolvepro/src/model.py.

Main Script: dms_main.py

The main script dms_main.py is used to run experiments with different combinations of grid search variables.

Usage

For dms, use the evolvepro environment:

conda activate evolvepro
python dms_main.py [arguments]

Arguments

  • --dataset_name: Name of the dataset
  • --experiment_name: Name of the experiment
  • --model_name: Name of the model used for embeddings
  • --embeddings_path: Path to the embeddings file
  • --labels_path: Path to the labels file
  • --num_simulations: Number of simulations for each parameter combination
  • --num_iterations: List of number of iterations (must be greater than 1)
  • --measured_var: Activity type to train on (options: activity, activity_scaled)
  • --learning_strategies: Type of learning strategy (options: random, topn2bottomn2, topn, dist)
  • --num_mutants_per_round: Number of mutants per round
  • --num_final_round_mutants: Number of mutants in final round
  • --first_round_strategies: Type of first round strategy (options: random, diverse_medoids, representative_hie)
  • --embedding_types: Types of embeddings to train on (options: embeddings, embeddings_pca)
  • --pca_components: Number of PCA components to use
  • --regression_types: Regression types (options: ridge, lasso, elasticnet, linear, neuralnet, randomforest, gradientboosting, knn, gp)
  • --embeddings_file_type: Type of embeddings file to read (options: csv, pt)
  • --output_dir: Output directory for grid search results
  • --embeddings_type_pt: Type of embeddings to use for .pt files (options: average, mutated, both)

Example SLURM Scripts

We provide two example SLURM scripts for different use cases:

  • esm_2_15B_full.sh: Performs a comprehensive grid search across multiple datasets and parameters.
  • esm2_15B.sh: Runs simulations with a specific set of optimal parameters across multiple datasets.

Running the Scripts

  1. Ensure you have the necessary dependencies installed and the correct conda environment activated.
  2. Submit the SLURM job using: 
    sbatch esm_2_15B_full.sh
    or 
    sbatch esm2_15B.sh
  3. The results will be saved in the specified output directory.

Note: These scripts make use of GNU Parallel to run multiple datasets in parallel. Make sure you have it installed and loaded in your environment.