When predicting if an infection is resistant or susceptible to a specific antibiotic, it is all too easy to think that the infection is homogeneous and, in fact, many bioinformatic variant callers encourage that point of view. Or, at best, you can subvert the format of, say, a variant call file (VCF) by using the functionality designed to report diploidy for reporting (up to) two mixed populations. (What plant geneticists do I have no idea).
Reality is likely messier, especially in a slow-growing persistent infection like tuberculosis and there have been previous studies suggesting that minor populations that are resistant to an antibiotic can come to dominate and should lead to a prediction of resistant.
In this free-to-read paper, Dr Alice Brankin shows how allowing just two or more reads that support one of the two most common resistance-conferring mutations to levofloxacin and moxifloxacin, leads to a significant improvement in the sensitivity of genetics-based resistance prediction with no significant drop in specificity.
This is important because the fluoroquinolones are present in several different drug regimes used to treat tuberculosis and brings their performance into line with other antibiotics (such as rifampicin and isoniazid) for which we believe we have a similar level of understanding of the mechanisms of resistance.