Generating synthetic genotypes using diffusion models

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Human genotype data is crucial for genetic research but poses privacy concerns, limiting public accessibility.
• Genome-wide association studies generate large volumes of sensitive genotype data.

Purpose of the Study:

• To introduce the first diffusion model for generating complete synthetic human genotypes.
• To enable the creation of DNA-level genomes from synthetic genotypes.
• To address the challenge of limited public availability of real human genotype data.

Main Methods:

• Development of a novel diffusion model for synthetic genotype generation.
• Validation of synthetic genotypes against real human genotypes using standard metrics.
• Training and evaluation of biomedical classifiers using both real and synthetic genotype data.

Main Results:

• Synthetic genotypes accurately mimic real human genotypes without direct reproduction.
• Biomedical classifiers trained with synthetic genotypes achieve accuracy comparable to those trained with real data.
• Augmenting real data with synthetic genotypes significantly improves classifier performance.

Conclusions:

• The developed diffusion model provides a method for generating realistic synthetic human genotypes.
• Synthetic genotypes offer a privacy-preserving alternative for data sharing and analysis.
• This technology is imperative for accelerating biomedical knowledge sharing in human genetics.