Processor made for AI speeds up genome assembly

Newswise – ITHACA, NY – A hardware accelerator originally developed for artificial intelligence operations is successfully accelerating the alignment of protein and DNA molecules, making the process up to 10 times faster than state-of-the-art methods.

This approach could make it more efficient to align protein sequences and DNA for genome assembly, which is a fundamental problem in computational biology.

Giulia Guidi, assistant professor of computer science at the Cornell Ann S. Bowers College of Computing and Information Science, led a study to test the performance of the accelerator, called an intelligence processing unit (IPU), using existing DNA and protein sequence data. The IPU speeds up the alignment process by providing more memory to speed up data movement – ​​a common delay.

“Sequence alignment is an extremely important and computationally intensive part of virtually any computational biology workload,” says Guidi. “It’s very common and is usually one of the bottlenecks in the calculation.”

The study, “Space Efficient Sequence Alignment for SRAM-Based Computing: conference, November 14. Max Xiaohang Zhao, also a former visiting scholar at Cornell, now at Charité Universitätsmedizin, is also co-first author.

With her research, Guidi wants to help scientists solve problems that they have not yet even tried because they require so much computing power. These complex problems require large-scale computation: assemblies of processors, memory, networking, and data storage that can handle large computing tasks.

Aligning sequences of DNA or proteins is one of these complex problems. When sequencing a genome, biologists end up with thousands or millions of short DNA sequences that have to be put together like a puzzle. They use an algorithm to identify pairs of sequences that overlap, and then connect the pairs.

In the past decade, scientists have turned to graphics processing units (GPUs) – originally developed to speed up graphics in video games – to speed up sequence alignment by performing calculations in parallel. With the development of IPUs for AI applications, Guidi and her colleagues wanted to know if they could use the new accelerators to address this problem.

“The need for large-scale computation is growing for many domain sciences because we are now so much better at generating data than ever before,” Guidi said. “Parallel computing has gone from being a luxury to being a non-negotiable.”

IPUs appealed to Guidi because they have significant on-device bandwidth for transferring data and can handle uneven and unpredictable workloads. X-Drop, a popular sequence alignment algorithm, has a very irregular computation pattern. If two sequences match, the algorithm requires many calculations to determine the correct alignment, but if they don’t match, the algorithm simply stops. GPUs struggle with these types of irregular calculations, but the IPU excelled.

When Guidi’s group collected sequences from the model organisms E. coli and C. elegans using the IPU, they achieved 10 times faster performance compared to a GPU, which spends too much time transferring data unnecessarily, and 4.65 times faster performance. than a central processing unit (CPU) on a supercomputer.

What currently limits the size of genomes that scientists can process is the number of available IPU and GPU devices, as well as the data transfer bandwidth between the host CPU and the hardware accelerator. There is a lot of memory on the IPU, but transferring the data from the host causes a major bottleneck.

The team helped address this problem by reducing the memory footprint of the X-Drop algorithm by 55 times. This allowed it to run on the IPU and reduce the amount of data transferred by the CPU. As a result, the system was able to perform larger comparisons and more sequence comparisons on the IPU, which helped balance the uneven workload.

“You can leverage the large memory bandwidth of the IPU, which allows you to speed up the entire processing,” Guidi said.

If vendors can upgrade the data transfer process between the CPU and the IPU and improve the software ecosystem, Guidi expects to be able to handle larger genomes on the same IPUs.

“The IPU could become the next GPU,” she said.

Other co-authors of the study include Johannes Langguth of the Simula Research Laboratory and Aydın Buluç of the Lawrence Berkeley National Laboratory.

-30-