Using machine learning to identify ancient RNA viruses in the human genome

A team of researchers affiliated with multiple institutions in Japan has used machine learning algorithms to help them identify ancient RNA virus remnants in the human genome.

Prior research has shown that when a person (or other animal) is infected with a virus, that virus can sometimes change the host's DNA by adding some of its own RNA. Other prior research has shown that ancient viruses that infected populations many years ago have sometimes left remnants of their RNA in the human genome.

Finding such remnants has proved challenging, however, due to the huge numbers of comparisons required for each suspected virus. In this new effort, the researchers used a machine- learning algorithm to help with the search.

To train the algorithm, the researchers used RNA from known non-retroviral endogenous RNA virus elements. After training, the researchers fine tuned their system to prevent as many false positives as possible. They then set it to work on the human genome and identified approximately 100 possibilities. After studying the possibilities, the researchers found that many of them were already known and many also fell below a threshold they had set to serve as a pass/fail option. That left them with just one possible unknown virus remnant.

The researchers then looked to see if the same remnant appeared in the genomes of other species such as marmosets and chimpanzees, and found that was, indeed, the case. That finding suggested the insertion had occurred at least 43 million years ago – before the species had diverged.