U-M team’s genome sequencing breakthrough featured in CACM Research Highlights

A team of researchers from the University of Michigan has been recognized for their innovative work in accelerating genome sequencing. Their 2023 paper titled “GenDP: A Framework of Dynamic Programming Acceleration for Genome Sequencing Analysis” was selected to be featured in the prestigious Research Highlights section of the Communications of the ACM (CACM).

GenDP was led by Yufeng Gu, a fifth-year PhD student in computer science and engineering (CSE) advised by CSE professor Reetu Das. Their work was done in collaboration with CSE professor Satish Narayanasamy, ECE professor David Blaauw, other CSE students, and researchers at Intel. The paper focuses on accelerating the computationally intensive tasks involved in genomic sequencing. With advances in genome sequencing technology generating increasing amounts of data, there is a critical need for more efficient computational frameworks to sift through and process this data. 

Dynamic programming (DP) is a fundamental building block in genomics, much like matrix multiplication is for machine learning. This creates strong demand for a general-purpose, programmable accelerator designed specifically for DP workloads. While Nvidia has recently advanced DP performance by adding DPX instructions to their Hopper GPUs, custom accelerators can deliver significantly higher gains, often achieving an order of magnitude better performance compared to GPUs and CPUs. The key is developing a specialized, programmable accelerator that can meet both current and future needs.

GenDP meets this challenge by introducing a programmable DP accelerator that delivers a remarkable increase in speed and efficiency, achieving a throughput and area speedup over general-purpose CPU-GPU systems by two orders of magnitude. It also comes within a 2.8-fold range of application-specific ASICs, which are optimized for specific tasks. These achievements mark GenDP as a significant step toward ensuring that genomic analysis computational power keeps pace with data generation.

“Our framework not only accelerates existing genomic sequencing tasks but does so in a way that is adaptable to future needs,” said Gu. “This flexibility is key, particularly as sequencing technologies and methods continue to evolve.”

The paper’s inclusion in CACM’s Research Highlights is a testament to the significance of the team’s innovations, as the process for selection is highly competitive. Papers are nominated from across ACM’s many conferences, with about 24 out of more than 10,000 papers ultimately chosen. The GenDP Research Highlight also includes a technical perspective by Yatish Turakhia, an independent expert. 

“This work exemplifies the impact that researchers at U-M are making in the field of computing,” said Das. “As genomic data continues to expand, advances like GenDP are essential for ensuring rapid and accurate genomic analysis.”