DC2_oligo

Accurate prediction of the oligomerization state of coiled-coil domains based on the representation of monomeric structures obtained with Colabfold

DC2_oligo is a program for predicting the oligomerization state of homooligomeric coiled-coil domains. It is based on a logistic regression model trained on the 217 embeddings generated by the AlphaFold2 multimer_v3 model for a single sequence.

The input for training is a data frame /tests/set5_homooligomers.csv and an embedding generated by the --save-representations option in colabfold (.npy) files. Only embeddings are required for prediction.

Example of running AlphaFold2:

colabfold_batch 0_monomer.csv . --num-models 5 --model-type alphafold2_multimer_v3 --num-recycle 5 --save-single-representations

Requirements and installation

1. Clone repository

git clone https://github.com/labstructbioinf/dc2_oligo

2. Create a virtual environment (using conda, for example) and install the requirements

conda create -n dc2_oligo
cd dc2_oligo
pip install .

3. Check if everything works with pytest

cd dc2_oligo
python -m pytest

Usage

python predict.py --cf_results DIR --save_csv STR

Argument	Description
--cf_results	Colabfold output directory with saved embeddings via --save-representations option
--save_csv	Save csv with input filename (optional)

python predict.py --cf_results tests/data/0 --save_csv testoutput.csv

Additional information

For best results, enter sequences that contain only coiled coil domain. You can easily identify such a domain with DeepCoil predictor. Please use the AlphaFold2 multimer embeddings (alphafold2_multimer_v3).