Section: New Results

Parallelism

Variant detection using processing-in-memory technology

Participants : Dominique Lavenier, Mohamed Moselhy.

The concept of Processing-In-Memory aims to dispatch the computer power near the data. Together with the UPMEM company (http://www.upmem.com/), which is currently developing a DRAM memory enhanced with computing units, we parallelized the detection of small mutations on the human genome. Traditionally, this process is split into 2 steps: a mapping step and a variant calling step. Here, thanks to the high processing power of this new type of memory, the mapping step can nearly be done at the disk transfer rate. In 2018, we define an ad-hoc data structure allowing the variant calling step to be performed simultaneously on the host processor. Basically, the two steps are overlapped in such a way that reads are mapped by packet. When a packet is mapped, the mapping results of the previous one dynamically feed the variant calling data structure. Performance evaluation on the FPGA UPMEM memory prototype indicates a very high speed-up (two orders of magnitude) compared with state-of-the-art software (specifically GATK).