Artificial Intelligence used to discover a potent antibiotic

 

BtoBio Innovation

Btobioinnovation.com

cid:image003.jpg@01D59339.822C7BF0

Author: Jean-Claude Muller, 穆卓Executive Editor at BtoBioInnovation  jcm9144@gmail.com

 

 

SPECIAL REPORT #20.6

 

Artificial Intelligence used to discover a potent antibiotic

 

 

Artificial Intelligence is an extremely powerful tool to analyse massive existing data, to identify unknown trends and even to decipher unexpected correlations. The technology is now being used in almost every single economic and industrial sector, including R&D and more recently in prospective discovery.

 

This report focusses on the use of artificial intelligence to identify a remarkable new antibiotic active against resistance bacteria. 

 

For many years it has been known that the massive use of antibiotics encourages the emergence of resistant pathogen strains. For most organisms, stressful environment promotes genetic changes. Bacteria exposed to sublethal levels of antibiotics produce reactive oxygen species which increases their genetic mutation, thus promoting the appearance of antibiotic-resistant genes and the development of the so called multidrug cross resistance (MDCR). Sublethal levels of antibiotics create a stressful environment which do not longer kill the bacteria, but trigger the emergence of new bacteria that are resistant to almost all existing antibiotics. Without new antibiotics, ten million lives around the world are on risk each year from infections according to a recent report. Finding a new antibiotic by the well-established and classical in vitro screening methodology can take up to ten years and often enough the new products are only slight variants of existing drugs and are limited to a narrow space of chemical diversity.

 

The new approach published in Cell by James J. Collins, Professor of Medical Engineering and Science at the MIT (Cambridge, USA) used a machine-learning algorithm to uncover potent antibiotic properties in other words an “in silico” screening. Using predictive computer models is not new, but until now, these models were not accurate enough.  “We wanted to develop a platform that would allow us to harness the power of artificial intelligence to usher in a new age of antibiotic drug discovery. Our approach revealed an amazing molecule which is arguably one of the most powerful antibiotics that has been discovered” said Collins. In previous models, molecules were represented as vectors reflecting the presence or absence of certain chemical moieties. With the use of new neural networks developed by Regina Barzilay, Professor of Electrical Engineering and Computer Science in MIT’s Computer Science and Artificial Laboratory (CSAIL), Collins could induce automatic representations, mapping a large number of molecules into continuous vectors which are used to predict their properties. In this specific case, Collins  had designed the model, on about 2,500 different molecules, including 1,700 FDA approved drugs and 700 natural products, to identify chemical features that makes molecules effective at killing E.coli.  With the trained model in hand, Collins tested the Brad Institute’s Drug Repurposing Hub, a library of 6,000 compounds, and found one molecule that was predicted to have strong antibacterial activity, and had a structure completely unrelated to any existing antibiotic. Using a different machine-learning model, the MIT researchers also showed that this identified molecule should have low toxicity to human cells. The molecule, which the group decided to call halicin, after the fictional artificial machine from “2001: A Space Odyssey”, was in fact well known since 2009 and had been investigated under the name SU-3327 as a possible diabetes treatment.

 

 

Halicin Chemical Structure

 

The potential new drug was synthesized and tested against dozens of bacterial strains collected from patients and showed indeed to be active against a significant number of MDCR strains including Clostridium difficile, Acinetobacter baumannii and Mycobacterium tuberculosis, but not against Pseudomonas aeruginosa, a lung pathogen difficult-to-treat.  The effectiveness of halicin was assessed in animal studies in using mice infected with a drug resistant Acinetobacter baumannii strain, a bacterium that has infected numerous American soldiers stationed in Iraq and Afghanistan. An application with halicin-containing ointment completely cleared the infections within 24 hours.  Very preliminary studies suggest that halicin disrupts the ability of bacteria to maintain an electrochemical gradient across their membranes, a vital gradient necessary to produce ATP and deliver energy to the cell. This type of killing mechanism is not very likely to produce rapid drug resistance. “When you are dealing with a molecule that likely associates with membrane components, a cell cannot necessarily acquire a single mutation or a couple of mutation to change the chemistry of the outer membrane. Mutations like to tend to be far more complex to acquire evolutionary” said Jonathan Stokes, one of the authors of the Cell paper.

To complete the previous study, the MIT researchers found that E.coli did not develop any resistance to halicin during a 30 day-treatment period whereas the same bacteria already started developing resistance within one to three days when exposed to ciprofloxacin, a second generation fluoroquinolinone.

After identifying halicin, Collins‘ group used the above described Artificial Intelligence models to screen more than 107 million molecules from the ZINC 15 data base, an online collection of more than 1.5 billion chemical compounds. Within less than three days they have identified 23 potential new candidates which were tested against five species of bacteria.  Eight new entities showed significant antibacterial activity and two of them were particularly potent.  Barzilay, now wants to use the deep learning algorithm to find more selective antibiotics, design new ones and optimize existing ones.

 

“This ground-breaking work signifies a paradigm shift in antibiotic discovery and indeed in drug discovery generally. Beyond in silico screens, this approach will allow deep learning at all stages of antibiotic development, from discovery to improved efficacy and toxicity through modifications and medicinal chemistry” said Roy Kishony, Professor of Biology and Computer Science Technion, the Israel Institute of Technology. “The work is really remarkable, their approach highlights the power of computer-aided drug discovery. It would be impossible to physically test over 100 million compounds for antibiotic activity” said Jacob Durrant, from the University of Pittsburgh.  The MIT team now plans to study halicin further and work with a biopharmaceutical industry partner or a non-profit organisation to rapidly develop the drugs in humans.

 

An burning question comes immediately to mind: can a similar approach be used to identify a new antiviral agent to combat coronavirus?

 

 

 

February 23, 2020

 

 

This document has been prepared by btobioinnovation and is provided to you for information purposes only.  The information contained in this document has been obtained from sources that btobioinnovation believes are reliable but btobioinnovation does not warrant that it is accurate or complete. The views presented in this document are those of btobioinnovation’s editor at the time of writing and are subject to change.  btobioinnovation has no obligation to update its opinions or the information in this document.

 

 

 

Share :
  •  
  •  
  •  
  •  
  •