Genomic Data Describe Population Structure but Struggle to Estimate Directed Connectivity Networks

The movement of individuals between fragmented populations, known as population connectivity, plays a crucial role in population persistence, genetic structure and, ultimately, the definition of a species distribution range. Advances in molecular biology have made genomic data readily accessible, even for non-model species, and they are increasingly used to estimate connectivity networks through diverse methods, including gene flow inference, clustering of individuals, and parentage assignment. Yet, no systematic study has evaluated the reliability of these different approaches in accurately estimating connectivity networks. To address this gap, we generated 3240 simulated SNP datasets based on a simple connectivity network of three virtual populations linked by unidirectional migration of known intensity. We then applied a range of analytical tools to these datasets to infer connectivity and evaluated their ability to accurately reconstruct the underlying connectivity network. We compared the accuracy of connectivity network estimates across methods and evaluated how four key parameters (population sizes, migration rates, and the sampling efforts of both individuals and SNPs) influenced their performance. Worryingly, we found that, across all methods, reliable estimates were achievable only under a narrow and largely unrealistic set of conditions: either complete individual sampling (100%) or moderate sampling (30%) coupled with low migration rates (< 10%). With lower sampling proportions (10% of individuals), as is often the case in current practice, none of the tested methods consistently recovered the true connectivity network. Among the tested methods, BA3-SNPs emerged as the most reliable overall, although other approaches may be preferable in cases of highly differentiated populations or very high proportions of sampled individuals. Crucially, this study highlights that increasing the number of SNPs to several thousands, as is often done in population genomic studies of non-model organisms, does not compensate for low proportions of sampled individuals, contrary to common assumptions. These findings raise a strong cautionary flag: without substantially high individual sampling efforts, there is a high risk of drawing misleading conclusions about connectivity, potentially leading to flawed ecological and management decisions. Our results therefore call for a careful reconsideration of current standards in the design of population genomic studies on non-model organisms aimed at inferring connectivity networks.