'Fingerprint' ML Technique Identifies Different Bacteria in Seconds

news.kaist.ac.kr

29 points·NotAWorkNick··2 comments

I guess the clinical relevance depends on how well you can tailor treatment based on it. AFAIK if docs don't know a) the specific bug and [1] b) its antibiotic resistance and c) the patient is seriously or potentially going to be seriously ill they start out with broad spectrum anyway.[2]

Since the fingerprint technique doesn't tell you antibiotic resistance the real benefit is probably identifying rare bugs early and to some degree avoiding broad-spectrum stuff and slowing down the sort of antibiotic resistance bacteria develop over a population over time.

[1] Docs can genrally guess the type of bug, or at least the class of bug, and even "narrow-spectrum" antibiotics can cover multiple classes

[2] For less serious infections docs go by likelihood tables and population features (ie-- UTIs are generally killed by X in this region of the world/country). If that treatment fails, they try another one.

I’m even less impressed because it looks like for their tests they used a rod and a cocci which are pretty different morphologically. If it could differentiate between group a and b strep or staph id be very impressed.

But also; many times it’s not just the antibiotic resistance but virulence that’s mediated by a plasmid and there are any visual distinctions. The effect would be biochemical in nature so a visual analysis would provide limited benefit.

Reply on news.ycombinator.com