As a specific type of structural variation, inversions are enjoying particular traction as a result of their established role in evolution. Using third-generation sequencing technology to predict inversions is growing in interest, but many such methods focus on improving sensitivity, giving rise to either too many false positives or very long running times. In this paper, we propose a new framework for inversion detection based on a combination of two novel theoretical models: rectangle clustering and representative rectangle prediction. This combination can automatically filter out false positive inversion predictions while retaining correct ones, leading to a method that has both high sensitivity and high positive prediction values (PPV). Further, this new framework can run very fast on available data. Our software can be freely obtained at https://github.com/UTbioinf/RigInv.

中文翻译：

---

使用矩形聚类和代表性矩形预测来预测局部反演。

作为一种特定类型的结构变化，反演因其在进化中的作用而受到特殊的关注。使用第三代测序技术来预测倒置的兴趣日益增长，但是许多此类方法专注于提高灵敏度，从而导致假阳性过多或运行时间非常长。在本文中，我们基于矩形聚类和代表性矩形预测这两种新颖的理论模型，提出了一种新的反演检测框架。这种组合可以自动滤除假阳性反演预测，同时保留正确的假阳性预测，从而导致一种方法既具有高灵敏度又具有较高的正预测值（PPV）。此外，此新框架可以在可用数据上非常快速地运行。