Machine Learning Approaches for the Prioritization of Genomic Variants Impacting Pre-mRNA Splicing
Machine Learning Approaches for the Prioritization of Genomic Variants Impacting Pre-mRNA Splicing
复制标题
DOI：
10.20944/preprints201911.0085.v1
复制
发表时间：
2019-11
期刊：
影响因子：
6
通讯作者：
Charlie F. Rowlands;D. Baralle;J. Ellingford
中科院分区：
文献类型：
--
作者：
Charlie F. Rowlands;D. Baralle;J. Ellingford
前体mRNA剪接的缺陷通常是孟德尔疾病的原因。尽管下一代测序的出现，允许更深入地了解患者的变异景观，但表征导致剪接缺陷的变异的能力并没有以同样的速度发展。为了解决这个问题，近年来利用机器学习方法的剪接预测工具数量急剧增加，这给临床遗传学家留下了大量的计算机分析选择。在这篇综述中，在基因组学和剪接分析的背景下介绍了机器学习的一些基本原理。然后使用一种关键的比较方法来描述七种最近的基于机器学习的剪接预测工具，揭示了高度多样化的方法和常见的警告。我们发现，尽管在生产特定和敏感的工具方面取得了很大进展，但仍有很大的空间用于预测变体对剪接的影响的个性化方法。这些方法可以提高诊断率，并支持改善患者护理。
Defects in pre-mRNA splicing are frequently a cause of Mendelian disease. Despite the advent of next-generation sequencing, allowing a deeper insight into a patient’s variant landscape, the ability to characterize variants causing splicing defects has not progressed with the same speed. To address this, recent years have seen a sharp spike in the number of splice prediction tools leveraging machine learning approaches, leaving clinical geneticists with a plethora of choices for in silico analysis. In this review, some basic principles of machine learning are introduced in the context of genomics and splicing analysis. A critical comparative approach is then used to describe seven recent machine learning-based splice prediction tools, revealing highly diverse approaches and common caveats. We find that, although great progress has been made in producing specific and sensitive tools, there is still much scope for personalized approaches to prediction of variant impact on splicing. Such approaches may increase diagnostic yields and underpin improvements to patient care.