Along with the 2 leading French research centers in genomics, UPMEM is now part of a French and EU public research initiative named GenoPIM, aiming at settling the base of PIM friendly genomics algorithms.
Strong of an inital €500K fund, the trio should propose the first portal for opensource PIM algorithms destined to guide all bioinformaticians hoping to leverage the benefits of PIM within their very computing hungry genomic application.
A bit of context:
High throughput DNA sequencing is now the main workforce for most genomics applications. They have already started to impact research and clinical use. Genome sequencing is now becoming a part of preventive and personalized medicine to identify, for example, genetic mutations for rare disease diagnosis, or to determine cancer subtypes to guide treatment options.
Currently genomics data are processed in energy-hungry bioinformatics centers, which necessitate the transfer of data via the internet, consuming substantial amounts of energy and wasting time. There is thus a need for fast, energy-efficient, and cost-efficient technologies to significantly reduce cost, computation time, and power consumption.
With GenoPIM, the consortium will leverage UPMEM PIM to enable such powerful edge computing with a focus on co-designing algorithms and data structures commonly used in genomics. The goal is to obtain the highest benefits in term of TCO.
The targeted applications are vast and will account the following:
- Sequence alignment: DNA or protein
- Variant calling on SNPs or on variant structures.
- Genome assembly: contigers, gap-filling, scaffolding
- Genome comparison
- Metagenomic analysis
UPMEM and INRIA have already partnered in the past on a paper: Variant Calling Parallelization on Processor-in-Memory Architecture, proving an acceleration at par with GPUs but for a fraction of the cost (x8) and energy consumption (x6)