Genetic code graphic

AI analyses gene variants for disease potential

Image credit: Dreamstime

An AI model developed by researchers from the University of Oxford and Harvard Medical School has demonstrated excellent capability for predicting the implications of human gene variants, identifying them as benign or causes of disease.

DNA mutation is a cardinal feature of biology. Small genetic variations – and the resulting proteins that build our cells – can lead to profound disruptions in physiological function, sometimes causing disease.

A handful of well-known genetic mutations and their associated conditions are well understood. However, dramatic leaps ahead in genome sequencing technology has not been followed with similarly rapid advances in the ability to interpret the meaning of the millions of genetic variations identified through human genome sequencing.

Harvard and Oxford researchers sought to make sense of these data by developing an AI tool called Eve (Evolutional model of Variant Effect). Eve uses machine learning to detect patterns of genetic variation across hundreds of thousands of non-human species and apply them to make predictions about the implications of variation in human genes.

In an analysis published in Nature, the researchers used Eve to assess 36 million protein sequences and 3,219 disease-associated genes across multiple species. The results suggest that 256,000 previously identified human gene variants – currently of unknown significance – should be reclassified as either benign or disease causing.

The researchers recommend using the tool to augment, not replace, current clinical models used to determine the meaning of gene variants. When used in combination with these approaches, Eve could boost the precision and accuracy of diagnosis, prognosis and treatment choice.

“Increasingly, people have access to sequencing their genomes, but making sense of the data is not always straightforward,” said Professor Debora Marks, a systems biology expert at Harvard Medical School who co-led the research. She and her colleagues emphasise that Eve is not a diagnostic test, but it can augment clinical tools to help make diagnoses, predict disease progression, and even choose treatment based on presence of certain mutations.

“We believe our approach can be used as an added tool in current clinical assessments and offers a powerful new way to reduce uncertainty and clarify decision-making, particularly in the clinical setting,” she added.

Their analysis showed that Eve outperformed other computational prediction models in predicting clinical effect, also scoring as high or better than current gold-standard high-throughput experiments for testing the effect of a mutation.

When it comes to interpreting the meaning of genetic variation, the stakes are extremely high. Reading a benign variation as disease causing can lead to erroneous diagnosis, further testing, anxiety, and even unnecessary medical interventions. Conversely, misinterpreting a disease-causing mutation in the genome as benign could do enormous harm by providing false reassurance when close observation, further testing, or intervention may be necessary.

Co-lead author and machine learning expert at Oxford, Professor Yarin Gal explained: “What we hope this approach will do is generate powerful data that can empower the clinicians on the frontlines to make the right diagnostic, prognostic, and treatment decisions.”

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles