3billion, a biotechnology company that makes a genome-based diagnosis on rare diseases using artificial intelligence, has developed a new deep-learning model, which detects more genetic mutations with twice higher probability than the existing model.

3billion announced Wednesday that their research results on “3Cnet,” the deep learning model that interprets genetic mutations, were published in Bioinformatics. 3Cnet is the first AI model that interprets genetic mutations applying knowledge transfer learning.

Its researchers created virtual genomic data to secure sufficient learning data and developed 3Cnet by applying a deep learning model so that it can use data on evolutionarily conserved human genome for learning.

As a result, 3Cnet showed 10 percent higher performance in interpreting currently known disease-causing genetic variants and 2.75 times higher performance in finding variants causing rare diseases than PrimateAI of Illumina, the largest company in the world in the genomic field.

Unlike the existing models that could interpret only missense variants among genetic mutations, 3Cnet also increased the number of interpretable mutation types by 69 percent, including frameshift, Indel, and Stop gain/loss.

“3Cnet has attained the highest level of performance among commercialized pathogenicity predictors, but we will not stop here but develop high-dimension AIs that can apply the latest natural language processing techniques and analyze the cause of genetic variant pathogenesis,” said Dr. Lee Kyoung-yeul of 3billion who led the research.

3billion uses advanced AI genetic diagnosis techniques, including 3Cnet, to interpret genetic mutations to diagnose rare disease patients. It is also growing rapidly in the global market of rare disease diagnosis based on Whole Exome Sequencing (WES) and World Geodetic System (WGS).