RNA-based regulation offers a promising alternative of protein-based transcriptional networks. However, designing synthetic riboregulators with desirable functionalities using arbitrary sequences remains challenging, due in part to insufficient exploration of RNA sequence-to-function landscapes. Here we report that CRISPR-Csy4 mediates a nearly all-or-none processing of precursor CRISPR RNAs (pre-crRNAs), by profiling Csy4 binding sites flanked by > 1 million random sequences. This represents an ideal sequence-to-function space for universal riboregulator designs. Lacking discernible sequence-structural commonality among processable pre-crRNAs, we trained a neural network for accurate classification (f1-score ≈ 0.93). Inspired by exhaustive probing of palindromic flanking sequences, we designed anti-CRISPR RNAs (acrRNAs) that suppress processing of pre-crRNAs via stem stacking. We validated machine-learning-guided designs with >30 functional pairs of acrRNAs and pre-crRNAs to achieve switch-like properties. This opens a wide range of plug-and-play applications tailored through pre-crRNA designs, and represents a programmable alternative to protein-based anti-CRISPRs.

中文翻译：

---

抗CRISPR RNA：通过对Csy4介导的RNA加工的深度学习来设计通用核糖调节剂

基于RNA的调控为基于蛋白质的转录网络提供了有希望的替代方法。然而，部分地由于对RNA序列至功能图谱的探索不足，使用任意序列设计具有所需功能的合成核糖调节剂仍然具有挑战性。在这里，我们报道了CRISPR-Csy4通过分析侧翼有超过1百万个随机序列的Csy4结合位点，介导了几乎全部或没有的前体CRISPR RNA（pre-crRNAs）处理。这代表了通用核糖化仪设计的理想功能序列。缺乏可加工的pre-crRNA之间可辨别的序列结构共性，我们训练了神经网络以进行准确分类（f1-score≈0.93）。受到详尽探测回文侧翼序列的启发，我们设计了抗CRISPR RNA（acrRNA），可通过茎堆积抑制pre-crRNA的加工。我们使用超过30个功能对的acrRNA和pre-crRNA对机器学习指导的设计进行了验证，以实现类似开关的特性。这打开了通过pre-crRNA设计量身定制的大量即插即用应用，并代表了基于蛋白质的抗CRISPR的可编程替代方案。