LoFreq-Star-Best-Practices

Best Practices for LoFreq-Star

Aligning your reads

LoFreq works with any mapper that produces BAM. However, we highly recommend to use BWA-MEM.

If your aligner does not produce mapping qualities or you don't trust them, then switch the use of mapping quality in LoFreq off (option -J for lofreq call). Two examples where this is likely appropriate are (1) BWASW, which seems to underestimate mapping qualities and (2) Bowtie, which reports only either 0 or 255, under some settings.

Some mappers have trouble placing the second read in paired-end sequencing. The result are so called orphaned reads. LoFreq ignores these by default (so does samtools) but you can include them if needed (option --use-orphan for lofreq call). Note, we have observed high number of orphaned reads in alignments created with Mosaik depending on the used parameters.

Note, we do not anymore recommend to remove non-uniquely mapped reads (i.e. those with mapping quality zero) as we did for previous version.

Pre-processing your BAM file:

We suggest to pre-process your BAM file by following GATK's best practices protocol. In a nutshell, this requires running the following steps: MarkDuplicates (Picard), indel realignment (GATK) and base quality recalibration (GATK).

If you are working on any other organism than human, GATK might complain about a missing vcf-file containing known variants (dbSNP). You can either use an empty vcf-file (only containing a header; even though that's cheating) or a vcf-file produced by running LoFreq on the raw BAM file (likely better alternative). Once done, run LoFreq again on the recalibrated file.

Under Construction: Suggested settings for certain scenarios

• BAQ, bonf, -S/-V, -l sens vs spec etc