I-Corps: AI for predicting polymer properties for biopolymer films

这个i-Corps项目的更广泛的影响/商业潜力是开发一个软件平台来预测生物聚合物等可再生材料的性能。在日益加剧的环境担忧中，亚马逊、沃尔玛和H&M等电子商务领军企业承诺到2030年减少塑料垃圾和碳排放。实现这一目标可能取决于可持续包装解决方案，该解决方案优先考虑生物降解性，而不是投资于回收工艺。此外，出于成本考虑，有必要将重点放在具有成本效益的包装用生物聚合物的生产上。拟议的技术提供了一种工具，能够定制生物聚合物的化学行为，可用于创建其应用所需的技术性能规范。拟议的软件平台可能具有影响多个行业的潜力，因为它能够发现和预测可再生材料的属性，包括消费电子产品、能量存储、生物燃料、食品提取和制药研究。该i-Corps项目基于模拟器的开发，该模拟器旨在最大限度地提高分子动力学模拟中的采样控制、效率和可扩展性。该模型使用强化学习控制策略作为归纳偏差，优化目标函数，如生成特定构象，从而最小化计算成本。将该技术应用于高分子链领域，与所研究的具有目标回转半径的化学高分子链构象相比，采样效率提高了40%，说明了机器学习控制策略在优化复杂模拟方面的有效性。最初的目标是将这一模型应用于由生物聚合物制成的包装膜。与类似的财产预测技术不同，该模型是基于物理的，这意味着不需要经验数据，并且它在传统计算机上运行，使其能够被更广泛的受众访问。基于RL的控制策略将被用来深入探索配置空间，以加强对模拟构象的控制，以进行力学性能预测，这是设计有效包装材料的关键方面。建议的技术可用于预测和调整生物聚合物的特定属性，如机械强度、灵活性或化学非反应性，这些对创建定制的包装解决方案至关重要。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the development of a software platform to predict the properties of renewable materials such as biopolymers. Amid rising environmental concerns, e-commerce leaders such as Amazon, Walmart, and H&M have committed to reducing plastic waste and carbon emissions by 2030. Achieving this goal may depend on sustainable packaging solutions that prioritize biodegradability over investment in recycling processes. In addition, cost considerations necessitate a focus on cost-effective production of biopolymers for packaging. The proposed technology provides a tool enabling the customization of the chemical behavior of biopolymers that may be used to create the technical performance specifications required for their applications. The proposed software platform may have the potential to impact multiple industries by enabling the discovery and prediction of renewable materials' properties including consumer electronics, energy storage, biofuels, food extraction, and pharmaceutical research.This I-Corps project is based on the development of simulator tailored to maximize sampling control, efficiency, and scalability in molecular dynamics simulations. The proposed model employs reinforcement learning control policies as an inductive bias, optimizing objective functions like generating specific conformations, which minimizes computational cost. Applied to the realm of polymer chains, the proposed technology has shown a 40% improvement in sampling efficiency over studied chemistry polymer chain conformations with a target radius of gyration, illustrating the power of machine learning control policies in optimizing complex simulations. The initial goal is to apply this model to packaging films made from biopolymers. Unlike similar technologies for property prediction, this model is physics-based, which means empirical data are not required, and it runs in conventional computers, making it accessible to a wider audience. RL-based control policies will be leveraged to explore the configuration space thoroughly to enhance control over the simulation's conformations for mechanical property prediction, which is a key aspect in designing effective packaging materials. The proposed technology may be used to predict and adjust specific properties of biopolymers, such as mechanical strength, flexibility, or chemical non-reactivity that are crucial for creating tailored packaging solutions.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.