UPVC performs small paired reads mapping (~150 nucleotides per read) onto a reference genome followed by variant calling on the Host CPU. UPVC takes advantage of the high PIM parallelism to complete the mapping stage. It consists in finding all potential positions of short reads on the map of the reference genome loaded in memory. The CPU relieved from the data heavy computation is available to perform advanced dynamic computing and complex tasks such as alignment with variant calling.
The input dataset is composed of an indexed human reference genome generated from a fasta text file, along with two fastq files containing the paired reads and some metadata. UPVC outputs a text file listing called variants in vcf format.
This program was developped by UPMEM team. Reach us at [email protected] if you would like more details about this implementation (workflow structure, benchmarks, etc.).
cmake -B build
cd build
make
First generate the index with:
make mrams
You can then run on the integration dataset with:
make run
make check
Or, if not running on upmem DIMMs, you can instead run the integration dataset with:
make run_simu
make check
Data-set must consist of 3 files:
<dataset_prefix>.fastathe reference genomee (or part of genomee)<dataset_prefix>_PE1.fastqthe PE1 of the input to compare to the reference<dataset_prefix>_PE2.fastqthe PE2 of the input to compare to the reference<reference_vcf>the reference vcf output to check the quality of the computation
Run once to create the MRAM for the reference genomee to compare to:
./<path_to_build>/host/upvc -i <dataset_prefix> -n <number_of_virtual_dpus_during_execution> -g index
Then run:
./<path_to_build>/host/upvc -i <dataset_prefix> -g map [-n <number_of_physical_dpus_available>]
If the number of physical dpus available is not specified, the program will try to alloc every dpus available at runtime.
Results are in <dataset_prefix>_upvc.vcf
To check the quality of the results use:
python3 <path_to_build>/../tests/compareVCF.py <reference_vcf> <dataset_prefix>_upvc.vcf