OrensteinLab / N-TIMP2--MMP9

Predicting the affinity landscape of N-TIMP2/MMP9_CAT by combining deep neural networks and deep mutational scans

Introduction

We present a novel approach using deep neural networks (DNN) trained on High-Throughput Sequencing (HTS) data for protein mutagenesis library affinity screens. In this study, we have focused on the experimental raw data from the N-terminal domain of the tissue inhibitor of metalloproteinases 2 (N-TIMP2) with the catalytic domain of matrix metalloproteinase 9 (MMP9_CAT). Our goal is to comprehensively measure Protein-Protein Interactions (PPIs) and accurately predict unobserved affinity-enhancing or affinity-reducing variants.

Purpose

This repository includes the code related to our research project aimed at predicting the affinity landscape of N-TIMP2/MMP9_CAT by combining deep neural networks and deep mutational scans

Data

The experimental raw data utilized in this study is derived from the N-TIMP2/MMP9_CAT complex. High-Throughput Sequencing (HTS) data has been used to train deep neural networks, enabling us to accurately predict unobserved affinity-enhancing or affinity-reducing variants.

Setup Environment Instructions (Windows)

Before you proceed with the setup, make sure to have Python and Anaconda installed on your system.

1. Download the Code Repository:

   • Visit the GitHub repository: https://github.com/OrensteinLab/N-TIMP2--MMP9/tree/main/Code
   • Download the contents of the "Code" folder.

2. Inside the "Data" Folder, Add Raw Data:

3. Create a Virtual Conda Environment:

   • Open a command prompt.

   • Navigate to the directory where you downloaded the "Code" repository.

   • Run the following command to create a virtual conda environment named "my_env" with Python 3.9.16 and the required modules:

     conda create --name my_env --file requirements.txt python=3.9.16
     

4. Activate the New Environment and Run the Script:

   • Activate the environment using the following command:

     conda activate my_env
     

   • Run the scripts according to the provided usage instructions.

Usage

1. Run the pre-processing script:

Execute the script pre-proccesing_to_raw_data.py in the presence of the raw data files located in the data folder.

 python pre-proccesing_to_raw_data.py

Upon completion, the script will generate two files in the data folder:

• all_variant_ala.csv: Contains all variants containing Alanine (Ala)

• all_variant_no_ala.csv: Contains all variants without Alanine (Ala)

2. Train models script:

While running the train_model.py script, you will be prompted to enter input representing the action you want to perform.

 python train_model.py

The code supports the following three options:

1- Use all the data to train the model without making any predictions.

2- Split the data into three sets: a training set (80%), a validation set (10%), and a test set (10%). Train the model using the training set and make predictions on the validation set.

3- Split the data into a training set (90%) and a test set (10%). Train the model using the training set and make predictions on the validation set.

• If you choose Option 1, the trained models will be saved in the folder.

• If you choose Option 2 or 3, prediction files will be generated at the end of the script execution.

3. Get predictions:

After training the model in the previous section: running python train_model.py using option number 1, you can now make your own predictions!

There are three scripts available for performing predictions:

3.1 Independent Dataset Prediction:

First save your independent dataset file in the data folder. The file should be in csv format and include the following columns: Variant, Ki, Ki ratio, log_ki_ratio

The train_inference_ki.py script removes variants from the training dataset matching those in the external dataset. Then, trains the models and makes predictions for these variants.

3.2 All Single Mutations Prediction:

The script inference_heatmap.py performs predictions for all single mutations.

3.3 A desired variant prediction

The script predict_variant.py can predict the log2 ER of any variant you want. Run the script and enter the 7 relevant positions of the variant you want to check. The script will return the predicted log2 ER of the variant.

For example, when running:

  python predict_variant.py
  Please insert the variant 7 positions sequence:SINSVHT

The output is:

  The predicted log2 ER of variant SINSVHT is 0.17626083