Adaptive Automated Scientific Laboratory

自适应自动化科学实验室

• 批准号: EP/M015661/2
• 负责人: Larisa Soldatova
• 金额: $ 6.94万
• 依托单位: Goldsmiths University of London
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FM015661%2F2
• 关键词: Adaptive; Automated; Scientific; Laboratory

Our proposal integrates the scientific method with 21st century automation technology, with the goal of making scientific discovery more efficient (cheaper, faster, better). A "Robot Scientist" is a physically implemented laboratory automation system that exploits techniques from the field of artificial intelligence to execute cycles of scientific experimentation. Our vision is that within 10 years many scientific discoveries will be made by teams of human and robot scientists, and that such collaborations between human and robot scientists will produce scientific knowledge more efficiently than either could alone. In this way the productivity of science will be increased, leading to societal benefits: better food security, better medicines, etc. The Physics Nobel Laureate Frank Wilczek has predicted that the best scientist in one hundred years time will be a machine. The proposed project aims to take that prediction several steps closer.We will develop the AdaLab (an Adaptive Automated Scientific Laboratory) framework for semi-automated and automated knowledge discovery by teams of human and robot scientists. This framework will integrate and advance a number of ICT methodologies: knowledge representation, ontology engineering, semantic technologies, machine learning, bioinformatics, and automated experimentation (robot scientists). We will evaluate the AdaLab framework on an important real-world application in cell biology with biomedical relevance to cancer and ageing. The core of AdaLab will be generic.The expected project outputs include:- An AdaLab demonstrated to be greater than 20% more efficient at discovering scientific knowledge (within a limited scientific domain) than human scientists alone.- A novel ontology for modelling uncertain knowledge that supports all aspects of the proposed AdaLab framework.- The first ever communication mechanism between human and robot scientists that standardises modes of communication, information exchange protocols, and the content of typical messages. - New machine learning methods for the generation and efficient testing of complex scientific hypotheses that are twice as efficient at selecting experiments as the best current methods.- A significant advance in the state-of-the-art in automating scientific discovery that demonstrates its scalability to problems an order of magnitude more complex than currently possible.- Novel biomedical knowledge about cell biology relevant to cancer and ageing. - A strengthened interdisciplinary research community that crosses the boundaries between multiple ICT disciplines, laboratory automation, and biology.All outputs produced by the project will be made publicly available by the end of the project.

我们的建议将科学方法与21世纪的自动化技术相结合，目标是使科学发现更有效率(更便宜、更快、更好)。“机器人科学家”是一种物理实现的实验室自动化系统，它利用人工智能领域的技术来执行科学实验的周期。我们的愿景是，在10年内，许多科学发现将由人类和机器人科学家组成的团队完成，人类和机器人科学家之间的这种合作将比任何一方单独创造科学知识的效率更高。通过这种方式，科学的生产力将被提高，从而产生社会效益：更好的食品安全，更好的药物等等。诺贝尔物理学奖获得者弗兰克·威尔切克预测，一百年后最好的科学家将是机器。我们将开发AdaLab(一个自适应自动化科学实验室)框架，用于人类和机器人科学家团队的半自动和自动化知识发现。这一框架将整合和推进若干信通技术方法：知识表示、本体工程、语义技术、机器学习、生物信息学和自动化实验(机器人科学家)。我们将评估AdaLab框架在细胞生物学中的重要现实应用，与癌症和衰老的生物医学相关性。AdaLab的核心将是通用的。预期的项目产出包括：-AdaLab被证明在发现科学知识(在有限的科学领域内)方面比单独使用人类科学家的效率高20%以上。-一种用于对不确定知识进行建模的新型本体，支持建议的AdaLab框架的所有方面。-人类和机器人科学家之间的第一种通信机制，使通信模式、信息交换协议和典型消息的内容标准化。-新的机器学习方法，用于产生和有效测试复杂的科学假设，这些方法在选择实验方面的效率是目前最好的方法的两倍。-在自动化科学发现方面的一项重大进步，表明其对比目前可能的问题更复杂的问题的可扩展性。-关于与癌症和衰老有关的细胞生物学的新生物医学知识。-加强跨学科研究社区，跨越多个信息和通信技术学科、实验室自动化和生物学之间的界限。项目产生的所有成果将在项目结束时公布。

项目成果

Automating Sciences: Philosophical and Social Dimensions

科学自动化：哲学和社会维度

• DOI: 10.1109/mts.2018.2795097
• 发表时间: 2018
• 期刊: IEEE Technology and Society Magazine
• 影响因子: 2.2
• 作者: King R

Multi-task learning with a natural metric for quantitative structure activity relationship learning.

具有自然度量的多任务学习，用于定量结构活动关系学习。

• DOI: 10.1186/s13321-019-0392-1
• 发表时间: 2019
• 期刊: Journal of cheminformatics
• 影响因子: 8.6
• 作者: Sadawi N