Table Formats

The following are detailed descriptions of the table formats used to encode amplicon alignments and SVs. Columns are ordered for ease of understanding the information they encode; the actual order of the columns in R data objects or text files may differ. Tables are produced as data.tables in R and saved as RDS files.

Primers

The primers table has exactly two rows with metadata about the two primers inferred to have generated the read data. If it is a F/R primer pair relative to the genome, the forward (F) primer is always primer 1 and the reverse (R) primer is primer 2. Otherwise primers are sorted according to position along the concatenated genome.

Edges

The edges table has one row per edge, i.e., a connection between two nodes, with an odd number of edge rows per sequenced read segment, in format A-[J-A…]-A. The columns are listed here as found in the final edges table, i.e., after splitting of reads into segments and other filtering actions. The edges table carries the most comprehensive information about each read and is accordingly the most complex table.

Edge columns clip5/3, score5/3, nBases5/3, start5/3, end5/3, and hasAdapter5/3 are used in the svPore pipeline for finding adapter sequences in reads. They should be ignored in svDJ tables. Edge columns node1/2IsPrimer and isChimeric were used for filtering the edges table prior to the final output. They have constant values and can also be ignored. Finally, column nStrands is not consistently used by svDJ and should be ignored.

Junctions

The junctions table has one row for every distinct junction sequence found in the edges table, after applying junction quality filters. Junction edges aggregated into the same junction row did not necessarily have the same sequence throughout the entire read (where base differences are most likely errors), but did have the same sequence at the junction itself, as defined by the combination of node1, node2, insertSize and, if insertSize > 0, jxnBases.

Networks

The networks table further aggregates the junctions table after the construction of junction networks. The junction rows aggregated into a given network are inferred to have arisen from the same source molecule but diverged due to PCR or sequencing errors. There may be some “collisions” where two independent source molecules fortuitously had a small edit distance and were inappropriately placed into the same network. This outcome is infrequent due to the consideration of junction coverage during network analysis, but when it happens will reduce the true molecular complexity of the source sample in the networks table.

However, especially in the case of nanopore sequencing with its higher error rate, it is also likely that some reads diverged from their parent source molecules because they acquired multiple errors. Such reads fail to be merged into their true network, falsely increasing the apparent molecular complexity of the source sample. Importantly, these often result in junction networks with low coverage that can be filtered against. They also likely arise from lower quality reads, and can be filtered against by option --min-alignment-identity, which demands that the alignments flanking a junction had a high percentage of bases that matched the reference genome (implying that the junction sequence itself was also high quality).

It is up to the user to decide whether junctions or networks give the most reliable quantification of the data, whether to trust low-coverage networks, how aggressively to set quality filters, etc. In general, we think the network analysis is robust such that the networks table provides the fairest assessment of detected junctions and their relative abundance.

---