Unexpected high accuracy of landscape genetics inference with convolutional neural networks

During the last decade convolutional neural networks (CNNs) have revolutionized the application of machine learning methods to classification tasks and object recognition. These procedures can summarize with great effectiveness image data in key features that allow to classify and predict with exceptional precision. Here we show for the first time how CNNs provide highly accurate predictions of small-scale genetic differentiation and diversity in a subterranean rodent from central Argentina. Using microsatellite genotypes and high resolution satellite imagery we trained a simple CNN which was able to predict local Fst and allele diversity accounting for more than 99% of their variation. When trained with changed landscape settings the CNN still highly accounted for ~60% of variation emerging as a promising tool for population and conservation genetics.