feat: Initial implementation of Molecular bar codes handling using AGeNT

[ericblanc20]

Prototype implementation of mapping data generated with Molecular BarCodes (MBC) or UMIs.

Background

The MBCs are typically used on FFPE data, where library complexity may be low. These data are also often compromised by FFPE or oxo-G artifacts which require careful analysis & filtration of somatic variants. Base-quality re-calibration (BQSR) should be used in these cases.

Design

The implementation has 5 steps:

1. Trimming the MBC sequence at the read's 5' end and inserting it in the read name for further processing.
2. Mapping the reads
3. Merging separate libraries
4. Marking the duplicates using the information in read names
5. Performing BQSR

Steps 1 & 2 must be done separately on separate libraries, in order to easily insert read groups information requires by BQSR. The separate bam files must be merged before marking duplicates.

Implementation

Because of the multiple operations required to produce the final result, I opted to create a meta sub-step, which creates a Snakefile which handles all necessary steps. This is similar to the parallel wrapper, except that the steps are not chunks of the same operation on smaller regions, but logically different operations.

Benefits

1. The wrapper creating and running the Snakefile creates a temporary directory where all the disk-intensive operations occur. This avoid cluttering the work/<tool>.<library> with large files which are not final results.
2. The code is relatively straightforward, and blends well the the other features from the step (coverage analysis, ...)
3. There is natural parallelisation of libraries sequenced on multiple lanes.

Drawbacks

1. The code (as it is now) is inflexible: it is hard to see how another MBC tool could be added, and the mapper must be either bwa or bwa-mem2 (as they share the same input parameters). It is also currently not possible to opt out of BQSR.
2. The notion of meta sub-step may go against the whole snappy design.
3. The MBC tool AGeNT is a commercial software from Agilent. I don't think it is available on Bioconda. Because of time pressure, I don't have the time to look for alternatives (such as umitools).

Notes

The current implementation must be viewed as a prototype. If the meta sub-step concept is deemed acceptable for snappy, I have considered a few options to improve on the current implementation and make it more flexible.

1. The BQSR can easily be put under user control.
2. The mapper wrappers could be modified to adapt their parameters to enable compatibility with the MBC tool (AGeNT requires the -C option (append FASTA/FASTQ comment to SAM output) to be set, add read groups, run on separate libraries).
3. The current wrapper could be abstract, with mbc-tool sprecific concrete classes.

However, in my opinion, all changes and improvements must be weighted against an undue complexification of the ngs_mapping step.

[coveralls]

coverage: 85.736% (-0.1%) from 85.866%
when pulling 11d5464 on 461-support-for-molecular-barcodes
into 4874074 on main.

[mbenary]

One quick question, otherwise looks good to me.

[mbenary]

Why are you using get for one input and [ ] for the other?

[ericblanc20]

Because the first is always present, and should raise an exception if not, while the second is optional, and with the empty list as default.