Detecting selection in low-coverage high-throughput sequencing data using principal component analysis