DBAT: Deep learning-based Bio-sequencing Analysis Toolkit

Version 0.2.0

Description

DBAT (Deep learning-based Bio-sequencing Analysis Toolkit) is a versatile toolkit for biological sequencing data analysis, consisting of two deep learning modules:

• deepPeak: A deep learning framework to predict sequencing tracks and enable cross-species prediction of mutant profiles.

• deepLoci: A CNN model to predict featured genomic loci based on omics data.

DBAT has been tested on a Linux system (Ubuntu 24.04.3, GCC 13.3.0, CUDA 12.9, Python 3.12.12)

Authors and Contact

• Authors: Xuan Ma, Bochao Liu
• Email: skyxma@tjnu.edu.cn

Installation

Step 1: Set up a Conda environment (recommended)

conda create -n dbat python=3.9
conda activate dbat

Step 2: Install DBAT

cd dbat
pip install -e .

Test Data

The test directory contains:

File	Description
ath.fa	Arabidopsis thaliana Chr2, Chr3, Chr4 sequences
osa.fa	Rice (Oryza sativa) Chr10 sequence
ath_h3k4.bw	Arabidopsis H3K4me3 ChIP-seq (SRR3709930)
ath_h3k9.bw	Arabidopsis H3K9me2 ChIP-seq (GSM5574912)
ath_wt.bw	Arabidopsis WT sRNA-seq (GSE179796)
ath_ddm.bw	Arabidopsis ddm1 mutant sRNA-seq (GSE179796)
osa_wt.bw	Rice WT sRNA-seq
ath_omics.csv	Integrated Arabidopsis omics data

Implementation

Navigate to the test directory:

cd test/

1. deepPeak predicts sequencing peaks (normal mode)

Train on Arabidopsis H3K4me3:

dbat deepPeak --mode norm \
              --action train \
              --genome data/ath.fa \
              --input data/ath_h3k4.bw \
              --output out1

Predict rice H3K4me3 track:

dbat deepPeak --mode norm \
              --action prediction \
              --model out1/Peak_model.pth \
              --stats out1/Peak_stats.npz \
              --genome data/osa.fa \
              --output out1

2. deepPeak conducts cross-species prediction of mutant profiles

Train on Arabidopsis WT and mutant:

dbat deepPeak --mode cross_spe \
              --action train \
              --genome data/ath.fa \
              --ref_wt data/ath_wt.bw \
              --ref_mut data/ath_ddm.bw \
              --output out2

Predict rice mutant profile:

dbat deepPeak --mode cross_spe \
              --action prediction \
              --model out2/Ref_model.pth \
              --stats out2/Ref_stats.npz \
              --genome data/osa.fa \
              --targ_wt data/osa_wt.bw \
              --output out2

3. deepLoci predicts featured genomic loci

dbat deepLoci --input data/ath_omics.csv --output out3

Additional Notes

1. deepPeak allows to input suppressive data:

dbat deepPeak --mode norm \
              --action train \
              --genome data/ath.fa \
              --input data/ath_h3k4.bw \
              --suppress data/ath_h3k9.bw \
              --output out2

2. Training parameters are adjustable: Modify settings (e.g. batch size) in config/cfg.yml as needed.

3. deepLoci input format: The input CSV file must include a header as: ID,feature,data,sample,pos1,pos2,...