EvoAI

EvoAI enables extreme compression and reconstruction of the protein sequence space

• FoldX
• Anaconda

1. download EvoAI

git clone https://github.com/Gonglab-THU/EvoAI.git
cd EvoAI

2. install FoldX 5 software and put FoldX into software folders

3. install Anaconda software

4. install Python packages from Anaconda

conda create -n evoai python=3.10
conda activate evoai

conda install pytorch cpuonly -c pytorch
pip install biopython
pip install click

5. We have calculated the features and prediction results of wild-type protein and some mutated proteins, which can be downloaded from the Zenodo website.

• Place the downloaded and decompressed contents of the wt_data.zip in the wt_data folder.
• Place the downloaded and decompressed contents of the mut_data.zip in the mut_data folder.

bash predict.sh -o ./mut_data/D32E,S56R,M112L,I123K,R124W,T181S

This study involves model predictions carried out in two parts: one using DMS information and the other using EvoScan anchors. These correspond to the two boxplots on the left and right in Figure 5C, respectively. The DMS information part relies on the zero-shot predictions from the GeoFitness model.

As a result, there are two folders in the source code corresponding to the respective predictions.

The folder contains the code and results related to predictions based on DMS information. The folder corresponds to the predictions using EvoScan anchors.

EvoAI enables extreme compression and reconstruction of the protein sequence space