We generated data from several technologies and applied this approach. We note the clear edge of single-molecule data from Pacific Biosciences, and that among the amplification-based technologies, data from Illumina had the lowest bias. However, our results represent a performance snapshot, and are exquisitely sensitive to the fine details of laboratory process (including library construction and sequencing), which we anticipate will continue to be improved. For example, we compared several methods of library construction and, within fixed protocols, noted assay variability arising from process fluctuation. We emphasize the ongoing importance of such process monitoring as a prerequisite for meaningful inferences about protocol improvements. Our experiences lead to the general conclusion that a platform's sequencing bias is not an immutable property. Although Ion Torrent's platform generally suffered from the most bias, reagent, protocol, and computational changes may lead to future improvements.
