SSBlazer is a pioneering tool that utilizes deep learning to predict Single Strand Break (SSB) sites based on genome-wide single-strand break sequencing data with single-nucleotide resolution.
Access the complete SSBlazer Manual for more details.
Follow these steps to set up the SSBlazer environment:
-
Clone the repository and navigate into the directory:
git clone https://github.com/sxu99/ssblazer.git cd ./ssblazer -
Create a new environment using the provided environment file:
conda env create -f environment.yml -
Activate the newly created environment:
conda activate ssblazer
To annotate Single-Strand Break (SSB) sites on a genome, use the genome_pred.py script. This script reads a FASTA file, predicts SSB sites, and outputs the predictions in BED format:
python genome_pred.py --in chr1.fa --out chr1_annotated.bed --batchsize 128
--in chr1.fa: Specifies the input FASTA file for the genome to be annotated.--out chr1_annotated.bed: Specifies the output BED file where the annotations will be written.--batchsize 128: Sets the batch size to 128 for processing. Adjust this value depending on your machine's resources.
Perform mutation prediction using the following command:
python mutation_pred.py --genome hg19 --chr chr14 --pos 73659501 --ref T --alt C
--genome hg19: Specifies the genome version.--chr chr14: Represents the chromosome of interest.--pos 73659501: Identifies the position of the mutation on the chosen chromosome.--ref 'T': Indicates the reference allele, i.e., the original nucleotide at the mutation site.--alt 'C': Designates the alternate allele, which is the mutated nucleotide.
Train a new model using your datasets:
python train_from_scratch.py --train train.csv --test test.csv
The model weights will be saved in the ./models directory. After the model is trained, you can use it to predict break sites on new data by loading the model weights.
For easier access and usage, check out the SSBlazer Web Server.