Probabilistic methods for quality improvement in high-throughput sequencing data