Design of immunoreceptor protein through the integration of machine learning and Bayesian inference

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K18003
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 62010:Life, health and medical informatics-related
• Research Institution: Aichi Cancer Center Research Institute
• Principal Investigator: Guo Zhongliang 愛知県がんセンター(研究所), システム解析学分野, 研究員 (20875819)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Keywords: タンパク質間相互作用 / 機械学習 / タンパク質大規模言語モデル / トポロジカルデータ解析 / タンパク質設計 / TCR

Outline of Final Research Achievements

Accurate protein-protein binding affinity prediction is essential for understanding protein function, designing new proteins for treating diseases. However, experimental measurement of protein binding affinity is time-consuming and expensive. In this study, we addressed the overlooked issues in current binding affinity prediction models in protein design. We developed a high-accuracy, fast prediction method that integrates information from both protein 3D structures and amino acid sequences using multimodal learning. The performance of our model on benchmark datasets showed a significant improvement over existing methods, with the Pearson correlation coefficient increasing from 0.684 to 0.904. Additionally, through model analysis, we confirmed the efficacy of multimodal learning in predicting protein binding affinity.

Free Research Field

機械学習

Academic Significance and Societal Importance of the Research Achievements

近年，ウイルスやがん細胞に結合する抗体またT細胞受容体の設計が注目され，臨床を含め，多くの研究が行われてきた．しかし，設計されたタンパク質とターゲット分子の結合能を実験で測定するには時間と費用がかかる．また，既存の結合能予測モデルの予測精度が実用化に至っていないことも十分に認識されていない．本研究課題で提案した高精度かつ高速な結合能予測モデルは，深層学習を用いたタンパク質設計手法と組み合わせることで，効率的にターゲット分子と結合するタンパク質を発見できることが期待される．タンパク質間相互作用はタンパク質機能の基礎であり，結合能を正確に予測することは生命現象の理解につながる重要なステップとなる．