A given sequencing machine assigns a 'certainty' to each base call in the form of a per-base quality score (FASTQ format).

I have reads from several machines aligned to the corresponding reference (BAM format or similar). Ignoring SNPs and misalignments (for simplicity), mismatches should represent sequencing errors.

I'd like to parse through the FASTQ and BAM file and evaluate how closely the per-base quality score reflects 'true' sequencing errors.

Is there an existing efficient, user-friendly, open source, well documented tool to do this? Failing that, is there a paper doing this that claims to have made the tool freely available?

Another method would be to look at bases corrected using kmer analysis or similar. Again, the per-base quality score can be compared to the 'true' sequencing errors.

Any tool to do that? I guess it could be baked into an existing error correction tool.

Do short read aligners actually make use of per-base quality score?

Many thanks,

  • $\begingroup$ Fastq Quality is already a -10log(Pe) Pe= Probability of error (en.wikipedia.org/wiki/FASTQ_format#Quality) . If you want to correct the errors, you'll need to assess multiple reads and then establish a protocol to propose a more fit answer. $\endgroup$
    – 3nrique0
    Jan 13, 2020 at 19:30

1 Answer 1


I think what you are looking for is GATK's Base Quality Score Recalibration. Their documentation is great so I won't rewrite it here.

BQSR stands for Base Quality Score Recalibration. In a nutshell, it is a data pre-processing step that detects systematic errors made by the sequencing machine when it estimates the accuracy of each base call.

  • $\begingroup$ That tool adjusts the quality (if I understand correctly), but what I want to do is actually assess the quality estimate from two different technologies. e.g. does platform A over estimate the quality of GC calls? (for example). $\endgroup$
    – Dan Bolser
    Jan 14, 2020 at 17:12
  • $\begingroup$ You can take the results of the same library run on two technologies and plot them against each other and see how far they are from the ideal line. The plots would be similar to what is in the post above and all the data needed should be generated by BQSR. $\endgroup$
    – Bioathlete
    Jan 15, 2020 at 3:21
  • $\begingroup$ Can you explain in detail (sorry for being dumb) $\endgroup$
    – Dan Bolser
    Jan 25, 2020 at 19:00
  • $\begingroup$ Part of the recalibration is the creation of the data for the recalibrated v calculated plots in section 4 in the documentation linked above. You can run that on the two technologies individually and compare how far they are from the perfect case. You can use the RMSE metric for a direct empirical comparison too $\endgroup$
    – Bioathlete
    Jan 30, 2020 at 23:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.