本文へジャンプ

成果報告書詳細
管理番号20170000000675
タイトル*平成28年度中間年報 次世代人工知能・ロボット中核技術開発 (次世代人工知能技術分野)大規模目的基礎研究・先端技術研究開発 人工知能と実験自動化ロボットを統合した次世代創薬プラットフォームの開発
公開日2018/3/8
報告書年度2016 - 2016
委託先名株式会社MOLCURE
プロジェクト番号P15009
部署名ロボット・AI部
和文要約
英文要約Abstract:
Antibodies are natural biological compounds with extremely promising applications in medical science for the therapy of a wide variety of diseases, such cancer and rheumatism. In recent years, drug candidate prediction using artificial intelligence (AI) is attracting a great deal of attention due to its potential to foster rapid breakthroughs in the field of drug discovery. However, AI technology requires a massive amount of input data to enable machine learning, and this data can only be gathered through carefully designed biological experiments. Presently, the lack of data suitable for AI applications is due to a severe lack of scientists with a double expertise in both the fields of experimental biology and AI. Our plan is to integrate AI and experimental automation, by building a robot that is able to carry out the entire cycle of antibody drug design, testing, and optimization. In short, we are developing a complete robotic system for antibody drug discovery, which is comprised of an AI component, the “Antibody Search Engine”, and assisted by a state-of-the-art robot able to carry out experiments in a high-throughput manner.

The project’s final goal is to implement the following continuous feedback cycle:
(1) Massive acquisition of experimental data using random antibody libraries;
(2) AI training to find determinant factors of efficient antibody design;
(3) Automation of experimental testing of the AI output results and feedback of these experimental results back to the AI to further increase performance.

For each of the above, the goals defined in the “Cutting-Edge Research” section were as follows. For (1) “Massive acquisition of experimental data”, we determine the antibody parameters that Antibody Search Engine should predict and the experimental conditions for each screening experiment. After that, we prepare a large-scale screening experiment using deep sequencing (NGS). In parallel with this, we develop protocols to efficiently produce random antibody libraries on a large scale. In (2) “AI training to find determinant factors of efficient antibody design”, we develop a software called “Eigen Kernel” that extracts relevant features from antibody sequences, and a first prototype of the Antibody Search Engine is created. Finally, in (3) “Automation of experimental testing of the AI output results” aims to complete a prototype model of the robotic system for experimental operation automation.

We have reached the following milestones in 2016:
(1) Through the investigation of clinical trial cases, we set 10 essential parameters for Antibody Search Engine predictions, and are currently in the process of implementing a large-scale screening experiment based on these parameters. For the deep sequencing (NGS) protocol, we developed a method to obtain DNA samples directly from a random phage library and succeeded in dramatically improving the quality of the data acquired from the screening experiments. Experiments were carried out as scheduled and sufficient learning data was obtained for a future test run of the Antibody Search Engine. In addition, through experiments using a random RNA-peptide complex library, we have established the foundation for creating various libraries in the next 3 years.
(2) The basic design of Eigen Kernel and Antibody Search Engine was completed. For Eigen Kernel, we developed informatics tools for extracting the sequence of the variable region of single-domain antibodies, and for assessing sequence feature importance based on the sequence properties itself, such as the amino-acid index and k-mer frequencies.
(3) We completed the prototype of an automatic liquid dispensing system, including experimental sample processing with a magnetic beads system. Operational tests were conducted in the presence of collaborators at the University of Tokyo, where a demo version is currently available.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る