AI trained on primate DNA can predict human genetic disease risk

2023-06-05

A group of international researchers have utilized a novel AI algorithm to analyze DNA data from primates, further revealing genetic variations that contribute to human diseases. Initially, scientists sequenced over 800 individual samples from 233 species of non-human primates across 16 families, ranging from lemurs to gorillas. To interpret this data, they developed a new algorithm called PrimateAI-3D. PrimateAI-3D is built upon a deep learning language architecture similar to ChatGPT, but it is designed to simulate genomic sequences instead of language sequences. The research team trained its parameters using natural selection by exposing it to mutations that have been eliminated in our primate relatives. Through this process, the algorithm learned to identify benign gene variations and, by elimination, identify mutations that may lead to diseases. Subsequently, scientists applied PrimateAI-3D to identify potential harmful mutations in humans using health records and genetic variation data from over 400 individuals in the UK Biobank project. They found that the algorithm demonstrated "significant improvement" in predicting genetic risks for common human diseases. This method claims to identify pathogenic mutations more accurately than existing technologies, which also addresses the genetic bias originating from European ancestors. Kyle Farh, the co-author of the study and Vice President of Artificial Intelligence at Illumina, stated, "Despite having 8 billion people, our genetic diversity still looks like a population of 10,000 common ancestors." He added, "It has been clear for a few years now that the information contained within the human genome sequencing is not enough to truly understand the human genome." Combining data from both human and non-human primates is crucial, especially considering that DNA shared between primates exceeds 90%. Illumina's research suggests that if a certain genetic variation is tolerated in natural selection among other primates, it is 99% unlikely to cause diseases in humans. The findings of this study can be used to support health research, such as assisting scientists in prioritizing variants that are most likely to pose risks to humans. They can also help protect populations of other primate species. "I think we're just scratching the surface," said Farh. "There's so much to learn here, and I think it's very romantic to understand our own species by learning from other species." The complete study has been published in the journal Science. Source: https://thenextweb.com/news/ai-trained-on-ape-dna-predicts-genetic-disease-risks-humans

RECENT AI NEWS

RECENT AI TOOLS