Uncovering footprints of natural selection through time-frequency analysis of genomic summary statistics

Natural selection leaves a spatial pattern along the genome, with a distortion in the haplotype distribution near the selected locus that becomes less prominent with increasing distance from the locus. Evaluating the spatial signal of a population-genetic summary statistic across the genome allows for patterns of natural selection to be distinguished from neutrality. Different summary statistics highlight diverse components of genetic variation and, therefore, considering the genomic spatial distribution of multiple summary statistics is expected to aid in uncovering subtle signatures of selection. In recent years, numerous methods have been devised that jointly consider genomic spatial distributions across summary statistics, utilizing both classical machine learning and contemporary deep learning architectures. However, better predictions may be attainable by improving the way in which features used as input to machine learning algorithms are extracted from these summary statistics. To achieve this goal, we apply three time-frequency analysis approaches (wavelet transform, multitaper spectral analysis, and S-transform) to summary statistic arrays. Each analysis method converts a one-dimensional summary statistic arrays to a two-dimensional image of spectral density or visual representation of time-frequency analysis, permitting the simultaneous assessment of temporal and spectral information. We use these images as input to convolutional neural networks and consider combining models across different time-frequency representation approaches through the ensemble stacking technique. Application of our modeling framework to data simulated from neutral and selective sweep scenarios reveals that it achieves almost perfect accuracy and power across a diverse set of evolutionary settings, including population size changes and test sets for which sweep strength, softness, and timing parameters were drawn from a wide range. Moreover, a scan of whole-genome sequencing of central European humans recapitulated previous well-established sweep candidates, as well as predicts novel cancer associated genes as sweeps with high support. Given that this modeling framework is also robust to missing data, we believe that it will represent a welcome addition to the population-genomic toolkit for learning about adaptive processes from genomic data.