Nanogirl Michelle Dickinson: Computer 'brain' models new drugs super-fast to kill superbugs

Bacterial resistance to antibiotics means that more than 700,000 people die each year from infections that could previously have been treated. Photo / Getty Images

COMMENT

Antibiotics are medicines that kill bacteria but many bacteria have mutated to become resistant to antibiotics making it harder to treat infections. This means that many ailments that would previously have been simple to treat are now potentially fatal. As the most dangerous drug-resistant bacteria threaten us all, new hope has come in the form of an artificial intelligence system inspired by a classic science-fiction movie.

Bacterial resistance to antibiotics means that more than 700,000 people around the world die each year from infections that could previously have been treated. The outlook looks bleaker when data shows that the discovery and regulatory approval of new antibiotics has slowed over the last few decades.

A new discovery published in the journal Cell could change that by using the power of artificial intelligence to help scientists rethink their fight with bacteria.

The researchers trained an artificial intelligence system to learn like a baby human brain would using something called a deep neural network. To do this they taught a computer algorithm the properties of molecules atom by atom in the same way that we would teach a child to learn words after teaching them individual letters. As the algorithm added more molecules to its brain it started to be able to predict how these molecules might interact with others. Previously, new antibiotics have been developed by modifying what we already know about current drugs and building on this. This new artificial intelligence system is unique as it has zero assumptions about previously created drugs so is not limited to patterns that have already been defined by humans.

The system was initially trained using a collection of 2335 molecules to spot any that inhibit the growth of the bacterium Escherichia coli. These molecules came from a library of hundreds of antibiotics as well as more than 800 natural products from plants and animals. After this training, the researchers used the system to screen more than 6000 molecules from a library called the Drug Repurposing Hub. Here they asked the "brain" to predict if any of these molecules that don't look like conventional antibiotics could be effective against E. coli. It found about 100 molecules and the researchers carried out physical testing on them. One of these molecules, which was already being investigated as a potential treatment for diabetes, turned out to be a potent antibiotic using an unconventional mechanism.

While most antibiotics work by blocking enzymes, this new molecule disrupts the flow of protons across a cell membrane. They called it halicin after the intelligent computer HAL from the 2001 film A Space Odyssey.

In initial animal tests, researchers found halicin to be active against a wide range of pathogens, including a pan-resistant bacteria called Acinetobacter baumannii. Pan-resistant means it is currently resistant to every standard antibiotic available. The animal tests also showed that halicin had low toxicity to the animal and even after 30 days the bacteria hadn't developed any resistance against it.

With so much initial success the researchers then gave the neural network a much bigger library to read - 107 million molecular structures. In only three days it was able to process them all and identified 23 potential antibiotic molecules. Physical testing showed that eight of these had antibacterial activity and two of them were potently active against a range of pathogens including an antibiotic-resistant strain of E. coli.

The ability to model new drugs in a computer brain in days compared to the physical testing of millions of molecules over decades shows the advancement that artificial intelligence has for the medical field. While clinical trials are still needed, computer brains have the potential to accelerate what humans have taken decades to do in only days.