A {hardware} accelerator initially developed for synthetic intelligence operations efficiently hurries up the alignment of protein and DNA molecules, making the method as much as 10 instances sooner than state-of-the-art strategies.
This strategy could make it extra environment friendly to align protein sequences and DNA for genome meeting, which is a basic downside in computational biology.
Giulia Guidi, assistant professor of laptop science within the Cornell Ann S. Bowers Faculty of Computing and Data Science, led a research to check the efficiency of the accelerator, known as an intelligence processing unit (IPU), utilizing present DNA and protein sequence knowledge. The IPU accelerates the alignment course of by offering extra reminiscence to hurry up knowledge motion – a standard holdup.
“Sequence alignment is an especially essential and compute-intensive a part of mainly any computational biology workload,” Guidi mentioned. “This can be very frequent and it’s often one of many bottlenecks of the computation.”
The research, “House Environment friendly Sequence Alignment for SRAM-Primarily based Computing: X-Drop on the Graphcore IPU,” can be offered by co-first creator Luk Burchard, a former visiting scholar at Cornell and doctoral scholar at Simula Analysis Laboratory, on the Supercomputing2023 convention, Nov. 14. Max Xiaohang Zhao, additionally a former visiting scholar at Cornell, now at Charité Universitätsmedizin, can be a co-first creator.
In her analysis, Guidi needs to assist scientists clear up issues they haven’t even tried but as a result of they require a lot computational energy. These advanced issues require large-scale computation – assemblages of processors, reminiscence, networks and knowledge storage that may deal with massive computing duties.
Aligning sequences of DNA or proteins is one in every of these advanced issues. When sequencing a genome, biologists find yourself with 1000’s or tens of millions of quick DNA sequences that have to be put collectively like a puzzle. They use an algorithm to establish pairs of sequences that overlap, after which hyperlink up the pairs.
Previously decade, scientists have turned to graphics processing models (GPUs) – initially developed to speed up graphics rendering in video video games – to hurry up sequence alignment by working calculations in parallel. With the event of IPUs for AI functions, Guidi and her colleagues wished to know if they might harness the brand new accelerators to sort out this downside.
“The necessity for large-scale computation is rising for a lot of area sciences as a result of we’re so a lot better at producing knowledge now than ever earlier than,” Guidi mentioned. “Parallel computing moved from being a luxurious to one thing that’s non-negotiable.”
IPUs attracted Guidi as a result of they’ve substantial on-device bandwidth for transferring knowledge and may deal with uneven and unpredictable workloads. X-Drop, a well-liked algorithm for aligning sequences, has a really irregular computation sample. When two sequences are a match, the algorithm requires a number of computation to find out the best alignment, however once they don’t match, the algorithm simply stops. GPUs battle with this sort of irregular computation, however the IPU excelled.
When Guidi’s group assembled sequences from the mannequin organisms E. coli and C. elegans with the assistance of the IPU, they achieved 10-times sooner efficiency in comparison with a GPU, which spends an excessive amount of time transferring knowledge unnecessarily, and 4.65-times sooner efficiency than a central processing unit (CPU) on a supercomputer.
Presently, what’s limiting the scale of the genomes scientists can course of is the variety of IPU and GPU units out there, in addition to the bandwidth for knowledge switch between the host CPU and the {hardware} accelerator. There’s a number of reminiscence on the IPU, however transferring the info from the host causes a significant bottleneck.
The staff helped to deal with this situation by shrinking the reminiscence footprint of the X-Drop algorithm by 55 instances. This enabled it to run on the IPU and scale back the quantity of knowledge transferred from the CPU. Because of this, the system might run bigger comparisons and carry out extra of the sequence comparisons on the IPU, which helped to steadiness the uneven workload.
”You’ll be able to exploit the IPU excessive reminiscence bandwidth, which lets you make the entire processing sooner,” Guidi mentioned.
If distributors can improve the info switch course of between the CPU and IPU, and enhance the software program ecosystem, Guidi expects that she might course of larger genomes on the identical IPUs.
“The IPU might turn out to be the subsequent GPU,” she mentioned.
Extra co-authors on the research embrace Johannes Langguth of the Simula Analysis Laboratory and Aydın Buluç of Lawrence Berkeley Nationwide Laboratory.
Patricia Waldron is a author for the Cornell Ann S. Bowers Faculty of Computing and Data Science.
