Hideki Yamaguchi
- yamaguchi.h.et.al [at] gmail.com
- hideki-yamaguchi.com
- Yokohama, Japan
Hi, this is the web site of Hideki Yamaguchi. I am a data scientist in a pharma, trying to develop innovative machine learning methods for efficient molecular design. My academic background is theoretical statistical physics (M.Sc., statistical-physical analyses on biological macromolecules) and computational biology (Ph.D., language model-based protein design).
Interests: Machine learning, protein engineering, drug discovery, software development, problem solving.
Research achievements
Publications (peer-reviewed)
- Matsushita, T., Kishimoto, S., Hara, K., Hashimoto, H., Yamaguchi, H., Saito, Y. & Watanabe, K. (2024). Functional Enhancement of Flavin-Containing Monooxygenase through Machine Learning Methodology. ACS Catalysis, 14(9), 6945–6951. doi: https://doi.org/10.1021/acscatal.4c00826
- Yamaguchi, H. & Saito, Y. (2023). Protein language models. JSBi Bioinformatics Review, 4(1), 52-67. doi: https://doi.org/10.11234/jsbibr.2023.1
- Ogawa, Y., Saito, Y., Yamaguchi, H., Katsuyama, Y., & Ohnishi, Y. (2023). Engineering the Substrate Specificity of Toluene Degrading Enzyme XylM Using Biosensor XylS and Machine Learning. ACS synthetic biology, 12(2), 572–582. https://doi.org/10.1021/acssynbio.2c00577
- Yamaguchi, H. & Saito, Y. (2022). EvoOpt: an MSA-guided, fully unsupervised sequence optimization pipeline for protein design. Machine Learning for Structural Biology Workshop, NeurIPS 2022. [paper]
- Yamaguchi, H., & Saito, Y. (2021). Evotuning protocols for Transformer-based variant effect prediction on multi-domain proteins. Briefings in bioinformatics, 22(6), bbab234. https://doi.org/10.1093/bib/bbab234
Presentations (peer-reviewed)
- H.Y.. Protein design via natural language. ‘Eight problems in bioinformatics’ workshop in IIBMP2023 (keynote speaker).
- H.Y. and Yutaka Saito. MSA-guided, data-efficient sequence optimization algorithm for protein design. IIBMP2022 (poster).
- H.Y. and Yutaka Saito. Zero-shot property prediction and sequence exploration by protein language models. IIBMP2021 (poster).
- H.Y. and Yutaka Saito. Evotuning protocols for Transformer-based variant effect prediction on multi-domain proteins. IIBMP2021 (oral).
- Tomohiro Oniduka, H.Y., and Yutaka Saito. Development of multi-task learning method for optimizing properties of antibody drug. IIBMP2021 (poster).
- Shuzo Fukunaga, H.Y., and Yutaka Saito. Improvement in variant effect prediction accuracy with an augmentation by fitness translocation. IIBMP2021 (poster).
- H.Y. and Yutaka Saito. Evotuning protocols for Transformer-based variant effect prediction on multi-domain proteins. ISMB/ECCB2021 Special Session Representation Learning in Biology (Invited lightning talk). [Recorded presentation]
- H.Y. and Yutaka Saito. Accurate prediction of variant effects by efficient incorporation of evolutionary information into Transformer-based deep learning. PSSJ2021 (poster).
- H.Y. and Yutaka Saito. Evolutionary training protocols for deep representation learning of multi-domain proteins. IIBMP2020 (poster).
- Shuzo Fukunaga, H.Y., and Yutaka Saito. Correlation analysis between directed and natural evolution of proteins using deep representation learning. IIBMP2020 (poster).
Career
Data scientist
In the research division, I am involved in developing impactful machine learning methods for efficient molecular design.
Research engineer
SyntheticGestalt K.K. is an early-stage company based in Tokyo and London. As a research engineer, I developed cloud-based machine learning systems for drug discovery and industrially-useful enzyme design. Graph neural networks or large protein language models were my primary toolkits.
Ph.D. student
As a Ph.D. student in the Department of Computational Biology and Medical Sciences (CBMS), Graduate School of Frontier Sciences, I studied protein language models and its appliactions to efficient protein design. I was also a visiting student in the Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology (AIST). Supervised by Dr. Yutaka Saito. Doctoral thesis title: “Data-efficient protein design based on protein language models”.
Senior algorithm engineer
PKSHA Technology Inc. is a leading machine learning company in Japan, span out from a machine intelligence lab in the University of Tokyo. In the company, I developed multiple algorithms for computer vision, data analysis, or chemoinformatics, based on deep learning and other ML techniques. My clients were a number of top-tier companies in automobile, surveillance, healthcare, pharma, and other industries. Python was the main tool (mostly PyTorch for deep learning but sometimes TensorFlow v1/v2, OpenCV, and PyData stack).