Nov. 16, 2021 -- Although 2 decades have passed since scientists completely mapped the human genome, finding out which human genetic variants are harmless and which ones lead to illness and disability remains a challenge.
A new artificial intelligence tool called EVE (evolutionary model of variant effect) has now used gene variants from thousands of animal species to predict which human genetic mutations are potentially harmful, scientists report inNature.
Researchers fed EVE data, which it processed following strict rules, adapting to what it learned as it went along. In this way, EVE sorted through data and identified patterns, changing course if new information called old patterns into question.
Investigators first trained EVE on a set of 250 million protein sequences representing more than 140,000 current and extinct species. They then tested EVE with more than 3,200 human genes with known disease associations. EVE passed the test by correctly predicting the disease-related significance of the genes.
The scientists then fed EVE a diet of 36 million sequences reflecting variants of the same genes. Each sequence was altered by only a single protein building block.
Based on patterns EVE detected, it classified 256,000 of these variant sequences as either harmless or disease-causing. The variants had previously been viewed as having no clear significance.
EVE isn’t a diagnostic tool. But its ability to sort gene variants based on potential for harm may aid efforts to identify causes of disease or disease progression, scientists suggest. Such information could help with making new screening tests or targeted therapies.
Down the line, EVE might also be a tool that doctors could use for accurate diagnosis and targeted prognosis and treatment, the researchers say. But that prospect remains a long way off.