Data Reduction in SARS-CoV-2 Amplicon Sequencing: Effects of Read Subsampling and Normalization on Consensus Quality, Lineage Assignment, and Computational Burden

To track how viruses change and spread, scientists rely on a powerful tool called next-generation sequencing. Imagine taking a photograph of a virus's entire genetic code, a long string of instructions that tells the virus how to behave. By reading these codes from many different patients, researchers can spot new variants, understand how the virus is moving through a community, and watch for dangerous mutations. However, the process of taking these genetic photos is not always efficient. The machines used to read the code often produce far more data than is strictly necessary, much like taking a photograph with a camera that captures every single grain of sand on a beach when you only need to see the horizon. This excess data creates a heavy burden for computers, slowing down the analysis and making it harder to get timely answers during a public health crisis.

A team of researchers in Utah set out to solve this problem of digital clutter. They focused on a specific type of genetic sequencing used for the SARS-CoV-2 virus, which relies on a method called amplicon sequencing. This technique works by copying small, specific chunks of the viral genome many times over so they can be read clearly. A known side effect of this copying process is that some chunks get copied much more than others, creating an uneven distribution of data. To ensure every part of the virus is seen, scientists often sequence the samples far beyond what is needed, generating massive files that take a long time to process. The researchers wanted to know if they could safely throw away some of this extra data to speed things up without losing the ability to identify the virus or detect new variants. They tested two different ways to reduce the data: one that simply picked random pieces to keep, and another that actively balanced the data to fix the uneven copying.

The study involved analyzing genetic data from 220 clinical samples taken from patients and 12 samples collected from wastewater, which contains a mix of viruses from many people in a community. The researchers took the original, massive datasets and reduced them to various smaller sizes, testing how well the computer could still do its job. They compared the results of two strategies. The first strategy, random subsampling, was like taking a bucket of mixed marbles and blindly scooping out a smaller handful. This method kept the original unevenness of the data but made the total pile smaller. The second strategy, read depth normalization, was more like a smart sorter that looked at the pile, identified the marbles that were too numerous, and removed the extras to create a perfectly balanced handful.

The results showed that both methods successfully reduced the time and computer power needed to analyze the samples, but they worked best for different goals. For the patient samples, where the goal was simply to identify which version of the virus was present, the random scooping method worked well as long as enough data remained. However, the smart sorting method was superior at creating a complete and accurate picture of the virus's genetic code. It allowed the researchers to get high-quality results with significantly less data than the random method required. Specifically, the smart sorting method produced reliable genetic maps at a data depth of 200 times coverage, whereas the random method needed more than 500 times coverage to achieve the same level of completeness. This means that for routine patient testing, using the smart sorting method could save a great deal of computing time while still providing the clear answers doctors and public health officials need.

The story changed, however, when the researchers looked at the wastewater samples. These samples are complex mixtures containing many different viral strains at once, and the goal is to figure out exactly how much of each strain is present. In this scenario, the researchers found that throwing away data was dangerous. When they reduced the amount of data too much, the computer struggled to tell the different strains apart. The accuracy of identifying which strains were present dropped sharply once the data fell below a certain level. The smart sorting method, while good at balancing the data, sometimes created false signals when the data was too sparse, making it look like there were more different strains than actually existed. The random method held up slightly better at very low data levels but still failed to give a reliable picture of the community's viral mix. The study concluded that while data reduction is a powerful tool for speeding up routine analysis, it should be used with extreme caution for environmental monitoring. For wastewater, keeping a high volume of data is essential to accurately untangle the complex mix of viruses circulating in a population.

Ultimately, this work provides a clear guide for how to handle the flood of genetic data generated during viral surveillance. It shows that there is no single best way to reduce data; the right choice depends entirely on what question is being asked. If the goal is to quickly identify a virus in a patient, a lightweight approach that trims the data is efficient and effective. If the goal is to build the most detailed possible genetic map, a method that corrects for copying biases is the better path. But when trying to understand the complex, mixed populations found in the environment, the researchers found that less data is not better. In those cases, the full volume of information is necessary to avoid missing the subtle details that tell the true story of how a virus is spreading. By matching the data reduction strategy to the specific task, public health laboratories can make their systems faster and more sustainable without sacrificing the accuracy needed to protect communities.