De novo development of small CRISPR-Cas proteins using artificial intelligence algorithms

摘要 大尺寸CRISPR Cas蛋白阻碍了CRISPR介导的基因的有效利用， 由于缺乏这些大尺寸蛋白质的递送策略， 到达他们的目标地点。为了寻求解决这一挑战的办法，据报告， 寻找自然界中较小的Cas蛋白。人工智能（AI）的最新成就 在生物领域的应用展示了人工智能在解决生物问题方面的力量， 带来了新的视角来应对这一挑战。在本申请中，我们提出了一个证明- 概念技术，以适应用于蛋白质结构预测和战略板的AI算法 游戏开发新的蛋白质设计工具，以尽量减少Cas蛋白质的大小，同时保持其 功能我们的目标是开发一种新的蛋白质设计技术，以优化蛋白质的大小 并开发出经过实验验证的人工mini-Cas蛋白。我们假设人造的迷你- 利用人工智能技术可以设计具有引导双链DNA切割功能的Cas蛋白。 为了验证这一假设，我们提出测试开发这种小型Cas设计技术的可行性 有两种方法。在目标1中，我们将使用基于序列的方法， 基于注意力的神经网络来设计mini-Cas蛋白。在目标2中，基于结构的半 无监督强化学习将用于技术开发。的最佳候选者 设计的mini-Cas蛋白将通过分子动力学模拟进行评估， 生物化学和基于细胞的测定。我们提出的技术将有很大的发展潜力 通过提供优化蛋白质大小的工具， CRISPR研究领域的范式从“从自然中寻找”到“在实验室中设计”， 提供第一个在生物化学和基于细胞的测定中验证的人工设计的Cas蛋白， 解决CRISPR交付挑战。

Abstract The large-sized CRISPR Cas proteins have hindered the effective use of CRISPR-mediated gene editing in human therapeutics due to the lack of delivery strategies for these large-sized proteins to reach their target locations. To seek solutions to this challenge, multiple efforts have been reported to search for the smaller-sized Cas proteins in nature. Recent successes in artificial intelligence (AI) application in the biological field demonstrated the power of AI in solving biological problems and brought a new perspective to address this challenge. In this application, we propose to develop a proof- of-concept technology to adapt the AI algorithms used in protein structure prediction and strategic board games to develop novel protein design tools to minimize the Cas protein size while maintaining its function. Our objectives are to develop a novel protein design technology to optimize the size of protein and develop experimentally-validated artificial mini-Cas proteins. We hypothesize that the artificial mini- Cas proteins with guided double-strand DNA cleavage function can be designed using AI technologies. To test this hypothesis, we propose to test the feasibility of developing this mini-Cas-design technology with two approaches. In Aim 1, we will use the sequence-based methods with the generative and attention-based neural networks to design mini-Cas proteins. In Aim 2, the structure-based semi- unsupervised reinforcement learning will be used for the technology development. The top candidates of designed mini-Cas protein will be evaluated by molecular dynamics simulations followed by the biochemical and cell-based assays. Our proposed technology will have the great potential to advance the technical field of protein design by providing tools to optimize the size of proteins, shifting the paradigm of the CRISPR research field from “searching from nature” to “designing in the lab”, and delivering the first artificially designed Cas proteins validated in biochemical and cell-based assays to address the CRISPR delivery challenge.