Rapid antibiotic-resistance predictions from genome sequence data for Staphylococcus aureus and Mycobacterium tuberculosis

What is it about?

Infections are generally caused by bacteria or viruses, and we treat with antibiotics - this dates back to Fleming discovering Penicillin. However bacteria can evolve "resistance" to drugs - their DNA mutates over time, and they pick up new abilities. We are now finding "drug resistant bacteria" (on which specific drugs do not work) are spreading. So, nowadays, if a patient has a worrying infection, doctors test the infection to see which drugs will work (mix the bug and drug "in a test tube", and see if it dies). The trouble is a test tube is not the same as a human body, and this also does not tell you anything about "which version" of the bacterium it is. In this paper, we show that by getting the DNA of an infection (In this case S. aureus or M. tuberculosis), you can predict which drugs will/not work. We do this for a large number of samples (about 4500 altogether) We also show for S. aureus that it's as accurate as the standard clinical lab methods, and about as fast as the hospital pipeline. We also show the computational work can be done with a simple drag-and-drop app that could run on a laptop - no need to employ very expensive techies.

Featured Image

Why is it important?

This paper shows a number of things. First - it shows that detection of known drug resistance-causing mutations and genes can be done swiftly in an automated fashion. Second, it looks carefully at the impact of "low frequency mutations" - in some sense the "minority population" of the bacteria. If just 1/100 bacteria are drug resistant - does it matter? We expect it does in vivo, but should we take it into consideration in predicting the in vitro results? Third it uses large sample sizes to measure error rates for different drugs.