Collaborative Research: IIBR: Innovation: Bioinformatics: Linking Chemical and Biological Space: Deep Learning and Experimentation for Property-Controlled Molecule Generation

一个深刻的问题是，什么形式承认期望的属性和行为，这是跨学科研究的基础。这个项目的重点是一个分子生物学实例。近年来，公开的小分子数据库的快速增长，引发了对硅分子设计和优化的深度学习处理的大量研究和兴趣。虽然许多现有的深度学习方法证明了它们产生化学有效分子的能力，但它们目前在为湿实验室研究提供信息的能力方面受到限制，这些研究旨在施加控制并回答以下问题：您的信息学模型能否产生受限于感兴趣的生物特性景观的这些特定区域的分子？模型、发现和数据将在科学界广泛传播。调查人员将共同指导各级学生。他们将自己的努力与机构的基础设施联系起来，以扩大其教育和推广活动的影响，并确保在这个项目中聚集的不同学科的不同学生的参与。该项目推进了属性控制分子生成。推动它的一个关键观点是，机器学习模型需要位于生物数据和知识中。研究活动分为三个重点：(1)开发能够纳入生物约束的可推广和可解释的模型；(2)适应小的、不完整的和有噪声的湿实验室数据；(3)在主动学习公式下整合计算和湿实验室查询。该项目在机器学习、人工智能、生成式人工智能和生物科学的界面上催化协同和创新工作，以解决分子生物学中长期存在的挑战，包括广泛和具体的季铵化合物（QACs）、小消毒剂抗菌化合物，这些领域的结构创新一直非常缺乏，耐药细菌是一个无法应对的威胁。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

A profound question that underlies much inquiry across scientific disciplines is that of what forms admit desired properties and behaviors. The focus of this project is on a molecular biology instantiation. The rapid growth of publicly-available small molecular databases has spawned much research and interest recently in deep learning treatments of in-silico molecule design and optimization. While many of the existing deep learning methods demonstrate their ability to generate chemically-valid molecules, they are currently limited in their ability to inform wet-laboratory studies aiming to exert control and answer the following question: can your informatics model generate molecules that are constrained to these specific regions of a landscape of biological properties of interest? Models, findings, and data will be disseminated broadly with the scientific community. The investigators will jointly mentor students of all levels. They connect their efforts with their institution’s infrastructures to broaden the impact of their educational and outreach activities and ensure the participation of diverse students across the various disciplines that come together in this project.This project advances property-controlled molecule generation. A key insight propelling it is that machine learning models need to be situated in biological data and knowledge. The research activities are organized in three thrusts: (1) developing generalizable and interpretable models capable of incorporating biological constraints, (2) accommodating small, incomplete, and noisy wet-laboratory data, and (3) integrating computation and wet-lab inquiry under an active learning formulation. The project catalyzes synergistic and innovative work at the interface of machine learning, AI, generative AI, and the biological sciences to address long-standing challenges in molecular biology both broadly and specifically on quaternary ammonium compounds (QACs), small disinfectant antimicrobial compounds, where structural innovation has been sorely lacking and resistant bacteria represent an uncountered threat.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.