Zero-shot prediction of mutation effects on protein function with multimodal deep representation learning <<<<<<< HEAD

Quick Start

As a prerequisite, you must have SE(3)-Transformers installed to use this repository.

Dependences

conda install --yes --file requirements.txt

Usage

Generate per-residue representations

python inference.py --task ec --outfile embeddings.h5

Calculate log-ratio heuristic under the constraints of both sequence and structure

python inference_dms.py --task ec --outfile fitness_prediction.h5

Zero-shot prediction of mutation effects

python predict_mutation_effects.py testdata/fitness_prediction.h5 

ProMEP-guided protein engineering

To guide protein engineering, users need to generate the 'fitness_prediction.h5' file following the above instructions and provide the raw sequence. Then run 1_dms_scanning.py to:

1. generate the virtual single-point saturation mutagenesis library
2. calculate fitness scores for all mutants
3. and rank them accordingly

examples

cd examples
python 1_dms_scanning.py

Protein mutants sorted by fitness score will be stored in 'dms_data/scanning-cas9.csv', while the fitness score for each mutant will be recorded in 'score_data/cas9-score.csv'.

=======

origin/main