ニュース

「Biophysics and Physicobiology」に Takuma Matsushita, Yutaka Saito による "Machine learning and AI in enzyme engineering: From sequence representation to de novo design" をJ-STAGEの早期公開版として掲載

2026年11月18日 学会誌

日本生物物理学会欧文誌[Biophysics and Physicobiology]に以下の論文が早期公開されました。

Takuma Matsushita, Yutaka Saito
"Machine learning and AI in enzyme engineering: From sequence representation to de novo design"

URL:https://doi.org/10.2142/biophysico.bppb-v24.s004


Large Image

Abstract
Enzymes are biological catalysts that combine high selectivity with high catalytic efficiency under mild conditions, yet natural enzymes lack stability under industrial conditions and activity toward non-natural substrates, and their functional modification has therefore been a long-standing challenge. Directed evolution depends on ultra-high-throughput assay systems, while rational design confronts combinatorial explosion. This review surveys the advances in machine learning and AI that are overcoming these limitations. We begin by tracing the evolution of numerical representations of sequence and structure, from one-hot encoding through CNNs and Transformers to structure-aware and multimodal protein language models. We then present representative achievements in supervised modification based on the DBTL cycle and active learning, in zero-shot prediction requiring no experimental data, in de novo design using ProteinMPNN and RFdiffusion, and in discriminative models that predict the function of unannotated proteins. We close by discussing the challenges ahead: accounting for dynamic structure, predicting comprehensive physicochemical properties, and automating the development cycle through self-driving labs.

URL: https://doi.org/10.2142/biophysico.bppb-v24.s004



前のページに戻る