The SpliceTransformer (SpTransformer) is a deep learning tool designed to predict tissue-specific splicing sites from pre-mRNA sequences.
2024.10.24 Our paper has been published in Nature Communications! Link We are continuously maintaining and improving our tool and associated web services. If you have any suggestions or encounter any issues, please do not hesitate to contact us via email or GitHub issues.
If you use the code or the data for your research, please cite our paper as follows:
@article{You2024,
author = {You, Ningyuan and Liu, Chang and Gu, Yuxin and Wang, Rong and Jia, Hanying and Zhang, Tianyun and Jiang, Song and Shi, Jinsong and Chen, Ming and Guan, Min-Xin and Sun, Siqi and Pei, Shanshan and Liu, Zhihong and Shen, Ning},
title = {{SpliceTransformer predicts tissue-specific splicing linked to human diseases}},
journal = {Nature Communications},
year = {2024},
volume = {15},
number = {1},
pages = {9129},
month = {oct},
doi = {10.1038/s41467-024-53088-6},
issn = {2041-1723},
url = {https://doi.org/10.1038/s41467-024-53088-6}
}
The program is developed in Python and the source code can be run directly.
However, the model weights file is too large to upload to GitHub. We support two ways to retrieve the repository.
Update 2024/11/01: Due to high download volumes, we have exceeded our Git-LFS bandwidth limit and have temporarily disabled Git-LFS access.
Clone the repository from GitHub first, then download the model weights from Google Drive and put them into the model/weights folder.
git clone https://github.com/ShenLab-Genomics/SpliceTransformer.gitDownload the model weights
Please see the Software Dependencies section.
We recommend running the software on a computer with a GPU. However, the lightweight example data can also be handled with CPU only.
This package is supported by Linux. The package has been tested on the following systems:
- CentOS 7
- Red Hat 4.8.5
It should be compatible with other common Linux systems.
- Python 3.8
- numpy
- pandas
- pytorch>=1.10.2
- gffutils>=0.11.0
- tqdm
- sinkhorn-transformer
- pyfaidx
- pyvcf3
- pyensembl
We suggest using Anaconda and pypi to install python packages. bioconda channel of Conda is recommended.
Example:
conda create -n spt-test python==3.8 numpy pandas gffutils tqdm pytorch pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 -c pytorch -c conda-forge -c bioconda
conda activate spt-test
pip install sinkhorn-transformer pyfaidx pyvcf3 pyensembl
The SpliceTransformer requires a genome assembly file and a genome annotation database file to locate genes and strands.
(1) Download the genome assembly file from Ensemble. https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/hg38.fa.gz
The uncompressed file should be placed at ./data/data_package/ and renamed into hg38.fa
(2) Download the genome annotation file from gencode.
https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_44/gencode.v44.annotation.gtf.gz
The file should be placed at ./data/data_package/ and renamed into hg38.annotation.gtf.gz
The default configuration is for hg38. However, other versions of annotation can also be used (the fasta files should contain chromosome labels like chr1 rather than 1).
Run sptransformer.py to predict mutation effects. The example output should be a .tsv table file containing the prediction.
python sptransformer.py --input data/example/input38.vcf --output data/example/output38.tsv --reference hg38At the first running of the script, a hint message will be printed to guide users to build .db file for the .gtf files.
Expected run time for the sample vcf file on a normal desktop computer is no more than 5 minutes.
The code snippets for analysis performed in the article are represented in tasks_annotate_mutations.py.
python tasks_annotate_mutations.py [task]The [task] should be replaced by task names recorded in the file. The path of input files should be updated in the code.
The code snippets for analysis performed during earlier revision processes are represented in
old_annotator_pytorch.pyandold_annotator_pytorch_2.py
Warning: Some tasks can not run directly because the source data are not included in the repository. Details about how to get the input files are described in corresponding section of paper.
The python class Annotator in sptransformer.py showed examples for usage of SpTransformer. The code is able to be modified for custom usage.
Run this command to generate example outputs as data/example/output38.csv:
python sptransformer.pyDetailed options：
python sptransformer.py -I input_file -O output_file --reference hg38 --vcf True --raw_score Falseinput_file: A .vcf or .csv file containing variants, e.g. data/example/input38.vcf
output_file: The output file, e.g. data/example/output38.csv
reference: The reference genome, either hg38 or hg19
vcf: Set this to False if the input file is a .csv file. The input .csv file should contain at least 4 columns: "CHROM POS REF ALT"
raw_score: - True: Output raw Δtissue usage scores - False: Output tissue specificity prediction. The 95% threshold (as described in the article) is used.
Note: We have observed that the model's output may exhibit minor discrepancies across different GPU hardware. There is a small probability that the predicted results for individual mutations may be inconsistent across devices, but the model's statistical performance on large-scale datasets remains unaffected.
The file custom_usage.py showed an example of getting raw outputs of SpTransformer model.
python custom_usage.pyThis project is covered under the Apache 2.0 License.