Key Takeaways
- A new AI model, popEVE, predicts the likelihood of genetic variants causing diseases, significantly aiding in diagnosing rare genetic conditions.
- popEVE successfully identified over 100 novel genetic alterations linked to previously undiagnosed diseases in around 30,000 patients.
- Collaborations are underway to validate popEVE for clinical use, with the potential to enhance treatment options and drug development for genetic disorders.
Advancing Genetic Diagnosis with popEVE
Recent advancements in genetic research have unveiled the complexity of human DNA, containing tens of thousands of variants that can impact protein formation. While many genetic changes exist, only a select few are known to cause diseases. This situation prompts a critical question: how can researchers pinpoint harmful variants amid numerous benign ones?
To address this challenge, scientists from Harvard Medical School have developed popEVE, an artificial intelligence model that scores genetic variants based on their potential to cause disease. As detailed in a study published in Nature Genetics, popEVE categorizes variants along a spectrum from benign to pathogenic and has successfully distinguished between variants that cause childhood and adult diseases.
popEVE has identified more than 100 new genetic alterations associated with rare diseases, which previously went undiagnosed. Debora Marks, a co-senior author of the study and professor at HMS, emphasized the model’s goal of providing a clinically relevant ranking of variants. This innovation aims to expedite and enhance the accuracy of diagnosing single-variant genetic diseases, especially for rare conditions, and could aid in discovering new drug targets for genetic disorders.
The development of popEVE builds on earlier work from the Marks Lab, which created EVE, a generative AI model that leveraged deep evolutionary data to understand mutation patterns. EVE, however, struggled to compare variants across different human genes effectively. To create popEVE, researchers integrated enhancements including a large-language protein model and human population data, allowing comparisons between variants from distinct genes.
In extensive tests, popEVE demonstrated its ability to correctly differentiate between benign and pathogenic variants, identify healthy individuals versus those with severe developmental disorders, and determine the severity and inheritance of conditions. Importantly, the model maintained accuracy across diverse genetic backgrounds, without showing bias in its predictions.
The researchers applied popEVE to a cohort of approximately 30,000 patients who were previously undiagnosed with severe developmental disorders. The model’s analysis provided diagnoses for about one-third of these cases, revealing variants linked to developmental disorders that had not been documented before. Notably, 25 of these variants have been independently confirmed by subsequent studies.
Looking ahead, the Marks Lab aims to transition popEVE into clinical settings, having made the model accessible via an online portal. Collaborations are ongoing with institutions such as Boston Children’s Hospital and the Children’s Hospital of Philadelphia, allowing clinicians to utilize popEVE for variant interpretation in patient cases. Initial feedback indicates that popEVE is already contributing to rare disease diagnoses.
Despite the promising results, further validation is necessary to ensure popEVE’s safety and efficacy before widespread clinical adoption. The integration of popEVE scores into existing databases like ProtVar and UniProt will enable global researchers to leverage this tool in their work.
Ultimately, Marks believes that prioritizing variants based on predicted disease severity through popEVE will enhance diagnosis rates and foster developments in treatment and drug discovery for genetic diseases.
The content above is a summary. For more details, see the source article.