A research team with Spanish participation creates an AI model for the diagnosis of rare diseases

Thousands of genetic variants coexist in a single patient, making it extremely difficult to determine which ones may be causing a disease. This is one of the major fields in which artificial intelligence is having a direct impact. In this context, this publication describes a significant improvement on a previous method.

The new model, called popEVE, allows the predicted severity of mutations to be compared across the entire genome, rather than for individual genes in isolation. To achieve this, it incorporates an analysis of the presence of mutations in both healthy human populations, although the available information is still limited, and in large collections of genomes from other species. The underlying reasoning is that mutations that are less frequent in these reference sets are more likely to be harmful and to cause a particular genetic disease. Although the principles are simple, the technical implementation is sophisticated, based on deep generative models and Gaussian processes.

In terms of medical application, the publication demonstrates the method's ability to detect mutations directly related to a disease with near-perfect accuracy—mutations that were not part of the model's training set. It is particularly relevant that it can do so without the need for genetic information from parents, reducing the requirement to sequence entire families, as is routinely done today. However, as it is a computational method, its reliability will need to be confirmed by other groups and with new cases, so its practical implementation will still require time for further validation by the community.

Two additional aspects stand out in this publication. It is a collaboration between a group at the Centre for Genomic Regulation (CRG) in Barcelona, where Mafalda Dias and Jonathan Frazer share leadership of their laboratory, and Debora Marks' laboratory at Harvard Medical School in Boston, where Mafalda and Jonathan completed their postdoctoral training. This situation is interesting in that it contradicts two ‘dogmas’ of the scientific system: on the one hand, it is considered inappropriate to share leadership of a group, and on the other, it is recommended that young scientists do not share publications with their mentors. It is clear that times are changing and criteria must be adapted to new realities.