Network analysis as a potential method detecting population structure

Abstract

Limited gene flow can divide a species into populations, forming population structure. Population structure is important for evolution and conservation so methods for detecting them were developed. With prior assumptions of population structure, Fst can be used to describe the degree of differentiation between each population. Later, structure and Discriminant Analysis and Principle Component (DAPC) overcame those prior assumptions and are widely used contemporarily. One approach called network analysis is widely used in physics and social science to study the patterns in complex relationships using similarities and dissimilarities. Population structure represents patterns in complex relationships, so network analysis should be able to detect population structure. I tested the ability of network analysis to detect population structure by comparing it to Structure and DAPC. I used simulated data of 4 populations, each containing 200 individuals’ genotypes at 15 loci, with migration rates among them varying from 0.001 to 0.1 migrants per generation. I predicted that network analysis would perform better than the other two methods. Contrary to my expectations, network analysis performed poorly overall, and did not detect any population structure correctly at migration rates greater than or equal to 0.01. At migration rates of 0.05 and 0.1, network analysis detected the correct number of populations just 20% of the time, with individual assignment error rates of 47% and 60%. Thus, network analyses do not appear to be a useful alternative to Structure and DAPC. However, only one network clustering method was tested, and therefore future studies could test if other such methods improve the performance of network analyses.