Limitation

• In theory, mhcflow can be applied to datasets generated from targeted and/or amplicon-based sequencing techniques that capture a large number of reads from HLA regions (often with many PCR duplicates). However, this application comes with a significant runtime cost during the realignment step, unless sufficient computational resources are available to efficiently split the workload into smaller tasks.

• mhcflow outputs the fished reads as uncompressed fastq files in the process, which can consume considerable disk space depending on the size of a sequencing library and the targeted coverage of the HLA regions, etc. This issue stems from the academia version of novoalign which does not support compressed fastq input. A future request may introduce a --gz option for compression if you have access to the commercial version.

• Starting with v0.2.0, individual components in mhcflow are no longer accessible via the command line as they were in v0.1.0. Enabling command-line execution for each component remains a potential future feature.

• The current method for splitting the realignment workload is rather naive and does not guarantee balanced distribution across tasks. This imbalance may result in some realignment tasks waiting for one particularly time-consuming workload to complete.

• The typing algorithm derived from polysolver (implemented by the mhctyper component) can be sensitive to fluctuations in the number of reads realigned. It reports only the alleles with the highest scores as the final results. Examining the top 3 or 5 alleles is recommended if the final results are unexpected, because even one single additional alignment might shift the score difference.