群体基因组学方法:从经典统计学到有监督学习

Population genomics:From classical statistics to supervised learning

群体遗传学的一个主要研究目标是理解突变、自然选择、遗传漂变、群体结构和数量变化等进化力量如何共同影响基因组中的遗传变异.通过分析DNA序列多态数据,可以推测曾经作用于基因组的各种力量,进而探讨生物演化的过程.近年来,随着第二代DNA测序技术的快速革新,群体遗传学进入了基因组学时代,相关的方法在不断发展,并可将群体基因组学方法分为经典统计学方法和新兴的机器学习方法.前者包括经典群体遗传学统计量、单一统计量或多统计量联合检测自然选择、群体历史与自然选择的联合估计以及基于溯祖树和祖先重组图的方法.后者主要基于有监督学习,为群体基因组时代的大数据分析带来了全新范式.本文从理论基础出发,全面回顾了群体基因组学方法发展变化的历程,着重介绍了该领域的最新进展,并就未来的发展方向进行了展望.

It is essential to understand how the patterns of genetic variation in organisms have been shaped by different evolutionary forces,such as mutation,natural selection,genetic drift,population structure,and population size change.In recent years,with the rapid innovation of next-generation sequencing technology,we are facing the new era of population genomics.The relevant population genomics methods can be classified as classical statistics and supervised learning.The classical statistics methods include many popular ones for detecting natural selection and inferring the parameters of demography,which are based on single or multiple combined statistics.The supervised learning methods may promise a new paradigm to make sense of large datasets in the genomic era.Here a brief introduction was first given on the important theory in population genomics.Then we overviewed the recent research progress in population genomics and shared our perspectives on its future development.