Researchers at Sun Yat-sen University and the University of Sydney built an AI model called LucaProt that reads both the sequence and the predicted shape of a key viral gene, letting it recognize distant relatives that sequence-matching tools would miss.
The team ran LucaProt across 10,487 environmental samples from soil, air, hot springs and deep-sea vents, not from any single tested animal host. It flagged 161,979 likely RNA virus species, about 70,458 never described before, including genomes far larger than typical RNA viruses.
Because the viruses came from free-living environmental metagenomes rather than a tested animal or human host, almost nothing is yet known about what, if anything, most of them infect. The tool is now being applied to other datasets, and the great majority of the 161,979 entries remain uncharacterized beyond their genetic sequence alone.
Read the original study
One AI search added tens of thousands of new viruses.
Source
- Hou, X., He, Y., Fang, P., Mei, S.-Q., Xu, Z., Wu, W.-C., Tian, J.-H., Zhang, S., Zeng, Z.-Y., Gou, Q.-Y., Xin, G.-Y., Le, S.-J., Xia, Y.-Y., Zhou, Y.-L., Hui, F.-M., Pan, Y.-F., Eden, J.-S., Yang, Z.-H., Han, C., . . . Shi, M. (2024). Using artificial intelligence to document the hidden RNA virosphere. Cell, 187(24), 6929–6942.e16. https://doi.org/10.1016/j.cell.2024.09.027