I-Corps: Structure Based Machine Learning Aided Protein Engineering

I-Corps项目的广泛影响/商业潜力涵盖化学制造、制药开发和农业生物技术。这项技术可以帮助蛋白质工程师选择一种候选蛋白质，使其对化学或环境扰动的抵抗力更强。在一个应用中，该技术可用于延长候选药物在开发过程中的半衰期。在另一个应用中，它可以用合适的酶化学物质取代有毒和浪费的化学物质。这些算法可能有助于推动目前无法获得的新型绿色化学物质，因为尽管必要的酶可以执行所需的催化作用，但它们不能保持足够长的折叠时间以用于商业用途。在第三个应用中，该技术可以帮助提高这些关键药物的保质期，从而扩大它们可能受益的人群。这个I-Corps项目利用机器学习来加速蛋白质工程。卷积神经网络最近已成为解决许多计算机视觉挑战的首选人工智能（AI）解决方案，但其生物应用仍然很少。蛋白质，一串折叠的氨基酸，驱动着大多数生物过程，正以指数方式结晶，以解决它们的三维结构。该项目通过训练3D卷积神经网络将这两种资源结合起来，该网络表征了20种氨基酸中每种氨基酸独特的化学环境。同样的神经网络可以预测最适合给定环境的氨基酸。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。

The broader impact/commercial potential of this I-Corps project spans chemical manufacturing, pharmaceutical development, and agricultural biotechnology. This technology helps protein engineers take a candidate protein and make it more robust to chemical or environmental perturbations. In one application, this technology could be used to extend the half-life of pharmaceutical candidates during development. In another application, it could enable substituting toxic and wasteful chemistries with suitable enzymatic chemistries. These algorithms may help drive new green chemistries presently inaccessible because although the necessary enzymes can perform a desired catalysis, they cannot stay folded long enough for commercial use. In a third application, the technology can help improve the shelf life of these critical drugs, thereby expanding the population they might benefit.This I-Corps project leverages machine learning to accelerate protein engineering. Convolutional neural networks have recently become the preferred artificial intelligence (AI) solution to a number of computer vision challenges, yet their biological applications remain scarce. Proteins, strings of folded amino acids that drive most biological processes, are exponentially being crystallized to solve their three-dimensional structure. This project combines these two resources by training a 3D convolutional neural network that characterizes chemical environments unique to each of the 20 amino acids. The same neural network can then predict the amino acid best fitting a given environment.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.