DBP-Finder

Introduction

DBP-Finder is an advanced deep learning tool designed for the precise identification of DNA-binding proteins (DBPs). Leveraging the power of state-of-the-art pretrained language models (PLMs), particularly Ankh, DBP-Finder accurately predicts DNA-binding proteins from protein sequence data. This tool empowers researchers to gain deeper insights into genetic regulation and protein function.

Installation

Environment Setup:

git clone --depth 1 https://github.com/GavrilenkoA/DBP-Finder.git
cd DBP-Finder
conda env create -f env.yaml
conda activate DBP-Finder

Data

Preparation training dataset

The training dataset is created using the following script, which processes positive and negative samples based on provided metadata:

python3 -m scripts.data.prepare_data --binders_path <binders.fasta> --non_binders_path <non_binders.fasta> --path_yml <path.yml> --output_path <output.csv>

Arguments:

1. --binders_path:

   • Type: str.
   • Default: "data/uniprot/go_0003677_swissprot.fasta"
   • Description: Specifies the file path to the FASTA file containing the positive samples (i.e., DNA-binding proteins, labeled with GO:0003677).

2. --non_binders_path:

   • Type: str.
   • Default: "data/uniprot/notgo_0003723_notgo_0003677_swissprot.fasta"
   • Description: Specifies the file path to the FASTA file containing the negative samples (i.e., proteins that are neither RNA-binding nor DNA-binding).

3. --path_yml:

   • Type: str.
   • Default: "data/processed/neg_samples_annot.yml"
   • Description: Specifies the file path to a YML file that includes metadata on whether any negative samples may bind to nucleic acids (GO:0003676). This file helps in further filtering or analyzing the negative samples.

4. --output_path:

   • Type: str.
   • Default: "data/embeddings/input_csv/train_p3.csv".
   • Description: Specifies the file path where the final processed dataset, including both positive and negative samples, will be saved as a CSV file.

Usage

This script performs inference to predict DNA-binding proteins (DBPs) from sequences provided in a FASTA Uniprot file using Ankh pretrained protein language model.

Command

To execute the script, use the following command:

python3 -m training-pLM.inference <FASTA_FILE> <OUTPUT_NAME> [--gpu <GPU_ID>]

Example

From the ROOT directory:

python3 -m training-pLM.inference data/fasta/Human-dna-binders.fasta Human-dna-binders --gpu 0

Output

The output will be saved in the data/prediction directory with the filename Human-dna-binders.csv

Input Requirements

Input sequences must be in FASTA UniProt format. Sequences should only contain canonical amino acids:

S, N, Y, L, R, Q, D, P, M, F, C, E, W, G, T, K, I, V, A, H

Sequences must have lengths between 50 and 1024 characters. Duplicate sequences and identifiers will be automatically removed.

Logs and Warnings

The script provides detailed logging to inform you about:

• Sequences with invalid characters that are removed.
• Sequences that do not meet length criteria.
• Duplicate sequences or identifiers that are filtered out.