brianhill11/deepsnp

Variant calling in genomic sequencing data using deep learning

deepSNP

Plan

• design fetures better reflecting technology
• better training data
• better deep learning method

Datasets

GIB

WXS/WGS

Simulated data

Simulate from chr1

to calculate number of reads (N) N=(cov * G)/rl

For chr17, cov=30, rl=200 2x100bp N=(30 * 83M)/ 200==8M reads to be simulated using this command

Using WGS

/u/scratch2/n/ngcrawfo/ERROR_CORRECTION/wgs_simulation/wgsim/wgsim -r 0.03 -R 0.005 -e 0.02 -1 100 -2 100 -A 0 -N 8000000 /u/home/s/serghei/project/Homo_sapiens/Ensembl/GRCh37/Sequence/Chromosomes/17.fa WGS_chr17_1.fastq WGS_chr17_2.fastq >log

The output of this command (written to log file) has this format:

• Col1: chromosome
• Col2: position
• Col3: original base
• Col4: new base (IUPAC codes indicate heterozygous)
• Col5: which genomic copy/haplotype

Reads are simulated here /u/home/s/serghei/scratch/WGS_deepSNP

Now i am mapping using BWA

bwa mem ~/project/Homo_sapiens/Ensembl/GRCh37/Sequence/Chromosomes/17.fa WGS_chr17_1.fastq WGS_chr17_2.fastq | samtools view -bS - > WGS_chr17.bam
qsub -cwd -V -N bwa -l h_data=16G,time=10:00:00 run_BWA.sh 
/u/home/s/serghei/scratch/WGS_deepSNP

To try

• Train on simulated apply on real?