Project Description

Faster DNA sequencing and
alignment promises to improve
accuracy, cut time and costs

DNA sequencing has come a long way in a relatively short period of time. The first method for determining DNA sequences was established by Ray Wu at Cornell University in 1970, according to Wikipedia. Second-generation methods were developed in the late 1990s to 2000. Today, DNA sequencing has become a key technology across sciences and is even becoming commonplace at the consumer level through companies such as 23andMe—and third generation sequencing methods are currently under development.
To better understand the future of DNA sequencing and sequence alignment, we asked Dr. Tajana Rosing for her perspective. Dr. Rosing is a professor, a holder of the Fatamico Endowed Chair, and a director of the System Energy Efficiency Lab (SEELab) at the University of California San Diego (UCSD). Rosing and her SEELab research team are taking a novel approach—DNA alignment using brain-inspired hyperdimensional computing—to a third-generation improvement to current methodologies.

What is brain-inspired hyperdimensional computing?

Brain-inspired hyperdimensional (HD) computing attempts to emulate human cognition by computing with hypervectors rather than the traditional numbers-based algorithms. HD computations are defined by patterns that mimic the activity of neurons, which is a more effective alternative to computing with numbers. HD computing can be applied to a vast number of learning problems across scientific fields.

How does this approach benefit DNA alignment and identification?

We think hyperdimensional computing can be harnessed to map DNA sequences into HD space for alignment, classification and recognition tasks. Simply put, hyperdimensional DNA sequencing, called HDNA, consists of an encoder and associative memory. The encoder maps DNA sequences, turning them into hypervectors and combining them together to build a single model representing each output class. These models are then stored in associative memory and used to compare and classify an unknown DNA sample by checking for similarities against the already modeled hypervectors. Essentially, these computer computations would work more like the human brain when doing the comparisons.

What is the benefit to this method?

Our experimental evaluation shows that HDNA can achieve 99.7% classification accuracy for an empirical dataset, which is 5.2% higher than current state-of-the-art DNA sequencing techniques for the same dataset. Moreover, HDNA can improve the execution time classification by more than 100x compared to today’s CPU-based implementations. Put simply, HDNA improves accuracy and reduces the time and costs.

From a practical standpoint, how could society benefit?

What can take weeks to figure out could drop down to one or two days. This is huge for all types of applications, but it could dramatically advance “precision medicine,” which benefits specific groups of patients often through using genetic or molecular profiling. This personalized approach represents a promising way to tackle particularly difficult diseases or rare diseases, which often have a genetic origin and can be especially challenging from a treatment perspective.
Ideally, genetic analysis offers the promise of more targeted treatments, catching diseases earlier and saving money for both consumers and society. While HDNA is still in the experimental stages, it may help turn those promises into reality over the next few years.
Dr. Tajana Šimunić Rosing’s research interests are in energy-efficient computing, cyberphysical and distributed systems, among other things. She recently headed the effort on SmartCities that was a part of DARPA and industry funded TerraSwarm center. She completed her Ph.D. in 2001 at Stanford University, concurrently with her master’s in engineering management. Her Ph.D. topic was Dynamic Management of Power Consumption. Prior to pursuing her Ph.D., she worked as a senior design engineer at Altera Corporation.
About Rubin Anders
For almost 20 years, Rubin Anders has delivered scientific experts to support high-stakes litigation. Our research team is comprised of scientists and attorneys uniquely positioned to understand our clients’ needs, and we leverage our global network of more than 20,000 scientists to reach the expert best positioned to support each client’s success.