Accelerating lifesaving research using AWS PCS with Tune Therapeutics

Tune has been working to develop innovative therapies for common diseases since 2020. One of its ambitious projects looked to uncover new therapeutic targets for liver-related diseases that can be debilitating or even terminal. “Currently, there aren’t a lot of effective treatments for many of these diseases,” says Harry Winters, senior software engineer at Tune. “So, the ability to halt disease progression, or potentially even reverse it, can make a huge impact on a lot of people’s lives.”

To do this, the company needed to analyze a massive, multimodal, human single-cell dataset. In the past, Tune primarily performed single-modal analysis—for example, gene expression analysis. For this project, however, Tune wanted to look at multiple modalities in the dataset simultaneously. Using single-cell data for the more advanced multimodal analysis was vital for uncovering new and potentially important therapeutic targets that researchers might otherwise miss. “Using single-cell analysis, we can separate the data into different cell populations and look at different gene expressions and genome confirmations in those populations,” says Winters. “Then, we can uncover unique characteristics and potential targets of not only the liver as a whole but also specific cell types within the liver.”

Because of the large amount of data involved—about 50 TB of raw data—the processing time would be extensive. “It was the largest single-cell multimodal dataset that we’ve ever had access to,” says Winters. During processing, the total size of the data would increase to 0.5–0.75 PB because of the creation of temporary data. In the past, for these types of datasets, Tune worked with a third-party cloud vendor. However, the vendor’s cloud lacked the scalability and elasticity needed to process a dataset of this size. Because of this limitation, Tune couldn’t run Cell Ranger, the software that it prefers for single-cell data processing.

Next, Tune tried to run the solution on a single instance. “We got the biggest machine that we could get, and that worked, but not fast enough,” says Winters. The company estimated that it would have taken up to 12 weeks to process the data, which wasn’t an option. “That type of timeline doesn’t fit with our milestones and how quickly we need to analyze data,” says Jason Dean, senior director of bioinformatics at Tune.

To process data more quickly and accelerate research, the company determined that it needed an HPC solution. However, Tune had specific criteria for the new solution—for example, working seamlessly with Cell Ranger and provisioning a Slurm cluster so that the company could parallelize its software runs. Slurm is a scalable cluster-management and job-scheduling system that intelligently balances workloads, and using it would help Cell Ranger connect with an HPC environment efficiently.

After researching different options, Tune selected AWS as the cloud provider for its solution and chose AWS PCS because it met all its selection criteria, including dynamic scalability and elasticity and ease of use. The managed service can simplify Slurm cluster operations and power many applications, including Cell Ranger. “We were looking for solutions that would help us run the Slurm cluster with minimal engineering overhead,” says Winters. “We had used HPC services on AWS before, and the managed nature of AWS PCS was appealing. There wasn’t really anything to manage. It just worked right out of the box.”

Using AWS PCS to power Cell Ranger, the company was able to improve its data processing time by four to six times, from an estimated 12 weeks using a single instance to 2–3 weeks—including developer time. “That made it possible to get readouts very quickly,” says Winters. “Using AWS PCS, we can make decisions far more rapidly, which is critical for a biotech company.”

Tune can now uncover novel targets faster, accelerating its analytics workflows. “One of the bottlenecks of using AI or machine learning in biotech is data processing,” says Dean. “The data engineering component is often one of the hardest parts of the whole AI life cycle in biotech. The new solution closes a big gap in that process.”

Tune looks forward to using AWS PCS in future research to take advantage of features unique to AWS PCS, using the managed service to expedite the company’s development and time to market. “I will reach for AWS PCS again anytime we have one of these groundbreaking single-cell datasets or need to accelerate a Cell Ranger run,” says Winters.

Using AWS PCS, the company completed the data processing within its target time frame. This frees up Tune’s scientists to develop new, potentially life-changing treatments for liver-related diseases. And importantly, the company has created a framework that its biological machine learning models can use for other applications to find novel targets in record time.

“Getting answers in a matter of weeks, instead of months, can ultimately make a huge impact on patients’ lives,” says Winters.