Sequence Analysis & Assembly with the LOGAN Framework

The sequencing revolution has enabled a broad range of biological investigations to be undertaken at dramatically reduced cost. However, limitations in the bioinformatic approaches used to analyse these datasets produce sub-optimal results. Reference-based approaches, typically using a set of linear reference molecules, are inherently biased and cannot handle divergent or entirely novel regions. The alternative, de novo assembly, requires massive datasets, is computationally challenging and typically produces highly fragmented results. Furthermore, the established assembly approaches are strongly tailored to specific kinds of sequence datasets, with the De Bruijn Graph (DBG) approach suitable for short reads, and the Overlap-Layout-Consensus (OLC) approach used for long reads. As a result, it is difficult to efficiently assemble hybrid datasets consisting of both short and long read data. It is also challenging to exploit existing reference sequences within an assembly-based approach. To address these problems, we have developed a new sequence analysis framework, LOGAN. This allows the combined analysis of reads from short and/or long read datasets, optionally supplemented with genome or pan-genome references, in a highly efficient data structure. This approach avoids both the information loss inherent in the fragmentation step of DBG assembly and the lack of scalability of the OLC approach, while allowing flexible integration of reference sequences. Since all input sequences are represented within a single data structure, it offers the potential to use statistically robust techniques on the complete dataset throughout the analysis, avoiding the crude heuristics typically used.