Did you know that the same algorithm that recognizes your face when you unlock your phone could find a new antibiotic for one of the most challenging pathogens out there? As unbelievable as that might sound, it is in fact true! And the concept behind both is one that has stirred quite the controversy in the modern media: artificial intelligence.
With the help of trained neural networks, researchers have recently found a new antibiotic that specifically targets Acinetobacter baumannii – a Gram negative pathogen that causes severe nosocomial infections in patients with a weakened immune system. Finding a new antibiotic for A. baumanii has proven to be difficult through conventional screening techniques, since this particular group of germs has the ability to develop resistance swiftly by incorporating DNA from their environment.
The ABCs of neural networks
But what are neural networks and what do they do? Essentially, they are algorithms that function like the human brain does, but better: they are faster learners than their muse. There are several layers of interconnected “neurons”: first layer of neurons receives input, then sends the output to the second layer and so on. Each layer is receptive for a specific characteristic. For example, if such a network were to analyse an image, the first layer would collect information regarding the dark and light spots, the second layer would try to recognize the shapes created by the sharp edges of the picture, etc. And this all applies to several aspects of our daily lives: facial recognition softwares, driverless cars or even text suggestions. And just as a neural network manages to process an image, it also manages to predict the antibiotic properties structurally new molecules have against A. baumanii.
How AI worked in our favour
The researchers trained the neural network to assess the activity of different molecules (off-patent drugs and synthethic molecules) that inhibit the growth of A. baumannii. Then, they used the collected data to train a binary classifier: an algorithm that places data in one of two categories. In this case, the categories for the binary classifier were: molecule that has antibiotic properties or molecule that does not have antibiotic properties. Using this algorithm the researchers predicted the antibacterial activity of compounds that were not initially used against the pathogen.
The assesed compounds were provided by the Drug Repurposing Hub – “an open-access repository of more than 6,000 compounds”, according to their website. In the past, researchers created those compounds with the intention of treating several other diseases. However, the molecules ended up out of use, for one reason or another. The Hub offers those structures to researchers willing to find them another purpose. The trained neural network processed the 6,000 compounds, assessed whether they had antibacterial properties and whether they were structurally similar to already known antibiotics that work against A. baumanii. In the end, the molecules with the highest activity against the pathogen underwent in vitro testing. Abaucin proved itself to be the most suitable compound. Abaucin was also tested on a wounded mouse model, where it proved to be remarkably effective.
Although this is an astonishing step forward, it is still important to mention that structural analogs of abaucin need to be developed for enhanced activity in vivo. With the help of AI, the field of medicine could evolve at a stunning pace, providing a cure for the now deadly nosocomial infections. If you found this article interesting, stay tuned and stay curious with us! You can read the entire article here.
About the author…
Hello! My name is Ilinca and I am a third year medical student that dabbles in a little bit of everything. I have an undying thirst for knowledge and a great talent for procrastination. Oh and I love hairless cats with all my heart!