Laboratory of Informatics of Molecular Functions

Visiting Associate Professor Yutaka SAITO
E-mail: yutaka.saito[at]
Lab HP


Key words: Machine learning, Bioinformatics, Protein engineering, Epigenome, RNA, Robotic biology

◆ Information technology accelerates life science research

  Even with the recent advances in experimental technologies, life science research still involves numerous trials-and-errors and reproducibility issues. We tackle these problems using machine learning and bioinformatics approaches. We also develop informatics methods for enhancing the applicability of experiment robots, aiming to establish "laboratory automation by experiment robots" as a future direction of life science research.

◆ Machine-learning-guided design of biomolecules: proteins, mRNAs, promoters

The development of functional proteins such as antibodies and enzymes involves trials-and-errors where a lowly-functional wild-type protein is to be improved by random mutations. To accelerate this process, we proposed a machine-learning method based on Bayesian optimization that predicts mutations for improving the protein function to the desired direction. In the collaborative work with experimental biologists, we demonstrated that our method successfully altered the green fluorescent protein (GFP) into the yellow fluorescent protein (YFP). Recently, we also develop methods to design mRNA codon sequences for improved translational efficiency, and to design promoter sequences for improved transcriptional activity.

◆ Omics data analysis: from DNA methylation to 3D genome structure

We develop bioinformatics methods for omics data analysis. Examples are: BPLA Kernel that predicts noncoding RNAs from sequence data, ComMet that detects differentially methylated regions from DNA methylation data (bisulfite-seq), and Cosearge that detects spatial co-localization of genomic elements from 3D genome structure data (HiC-seq). We also apply these methods in the collaborative work with experimental biologists: BPLA Kernel was used to find snoRNAs in the worm genome, and ComMet was used to analyze epigenomic changes in fat differentiation. Recently, we also study DNA methylation in industrially-important microorganisms.

◆ Robotic biology: towards laboratory automation by experiment robots

Experimental technologies in life science often involve reproducibility issues depending on persons who conduct experiments and/or laboratories where experiments are conducted. These problems could be resolved by experiment robots that can conduct exactly the same procedure repeatedly based on an electronically-described experimental protocol. However, current experiment robots have limitations e.g. they only support a limited set of pre-defined movements, hindering their applications. We are developing informatics methods for enhancing the applicability of experiment robots using "Maholo" installed in AIST as a model case (right figure). For example, we plan to develop "human-robot protocol translation" method that enables to transform a human experimental protocol to the equivalent robot protocol using a pre-defined set of movements.

Research environment: Students can select different research styles including (1) develop new machine learning and/or bioinformatics methods, (2) analyze publicly available data for new biological findings, and (3) analyze data in a collaborative work with experimental biologists. The laboratory location is Artificial Intelligence Research Center, AIST (Tokyo, Odaiba).


The University of Tokyo
Graduate School of Frontier Sciences, The University of Tokyo

Page Top