News Releases & Research Results A highly accurate method for genome sequence estimation with privacy protection ensured - A novel deep learning-based method -

The results of research conducted as the Tohoku Medical Megabank Project by the Tohoku University Tohoku Medical Megabank Organization.

• A novel genotype imputation method RNN-IMP (Recurrent Neural Network - IMPutation) was developed without the use of whole-genome data (reference panel) collected from a large population, thereby eliminating privacy concerns and allowing the accurate estimation of whole-genome sequences as with the conventional methods.
• The RNN-IMP method was intended for genotype imputation* by utilizing deep learning technology with numerical parameter data with which personal identification is difficult. This method should allow highly accurate genotype imputation, which could not be easily achieved with conventional mathematical models at many research institutions.
  *A technique for estimating unobserved gene mutation data from observed ones.

This program was conducted with the support of the Platform Program for Promotion of Genome Medicine by AMED.

The results were published online in the British scientific journal PLOS Computational Biology on October 2.

Kojima K., et al. A genotype imputation method for de-identified haplotype reference information by using recurrent neural network PLOS Computational Biology
DOI: 10.1371/journal.pcbi.1008207