variant effect prediction using SpliceAI

SpliceAI wrapper for retraining SpliceAI and variant effect prediction

Create environment as -

python -m venv spliceai
source spliceai/bin/activate
pip install -r requirements.txt

We will require bedtools for grabing sequences to create the dataset that we can refer here for installing it.

Firstly, Update the constants.py file:

• ref_genome: path of the genome.fa file (hg19/GRCh37) or (hg38.fa)
• splice_table: path for reference splicing sequences (canonical_dataset.txt for hg19 and hg38V46_splice_table.txt for hg38)
• sequence: for sequence name
• version: is used for naming the file

Then, download the appropriate genome FASTA file for your dataset.

cd data

For GRCh37/hg19

!wget http://hgdownload.cse.ucsc.edu/goldenPath/hg19/bigZips/hg19.fa.gz

Or for GRCh38/hg38

!wget http://hgdownload.cse.ucsc.edu/goldenPath/hg38/bigZips/hg38.fa.gz

Finally, use the following commands for data preprocessing:

cd data/
./grab_sequence.sh

python create_datafile.py train all
python create_datafile.py test 0

python create_dataset.py train all
python create_dataset.py test 0

use the following commands for training the models:

make sure constants.py has correct version set as hg19 or hg38

cd spliceai-training/
qsub script_train.sh 10000 1
qsub script_train.sh 10000 2
qsub script_train.sh 10000 3
qsub script_train.sh 10000 4
qsub script_train.sh 10000 5

use the following commands for training the models:

qsub script_test.sh 10000

If we want spliceai predictions using their original tools on some sequences (stored as fasta files with additional context of say 5k for SpliceAI-10k on each side of the sequence) we run the following command-

python spliceai-training/get_scores.py

The above script assumes all SpliceAI weights are stored in Models/pre-trained directory , sequences to be evaluated are present in data/sequences directory and the output results are to be stored in spliceai-training/sequence_output_predictions directory. These can be changed by changing INPUT_DIR and OUTPUT_DIR variables in the script.