Compression of DNA sequencing data

Jan Voges

Abstract: With the release of the latest generations of sequencing machines, the cost of sequencing a whole human genome has dropped to less than US$1,000. The potential applications in several fields lead to the forecast that the amount of DNA sequencing data will soon surpass the volume of other types of data, such as video data. In this dissertation, we present novel data compression technologies with the aim of enhancing storage, transmission, and processing of DNA sequencing data. The first contribution in this dissertation is a method for the compression of aligned reads, i.e., read-out sequence fragments that have been aligned to a reference sequence. The method improves compression by implicitly assembling local parts of the underlying sequences. Compared to the state of the art, our method achieves the best trade-off between memory usage and compressed size. Our second contribution is a method for the quantization and compression of quality scores, i.e., values that quantify the error probability of each read-out base. Specifically, we propose two Bayesian models that are used to precisely control the quantization. With our method it is possible to compress the data down to 0.15 bit per quality score. Notably, we can recommend a particular parametrization for one of our models which—by removing noise from the data as a side effect—does not lead to any degradation in the distortion metric. This parametrization achieves an average rate of 0.45 bit per quality score. The third contribution is the first implementation of an entropy codec compliant to MPEG-G. We show that, compared to the state of the art, our method achieves the best compression ranks on average, and that adding our method to CRAM would be beneficial both in terms of achievable compression and speed. Finally, we provide an overview of the standardization landscape, and in particular of MPEG-G, in which our contributions have been integrated.