Antibiotic resistance: Is artificial intelligence our only saviour?
“10 million people worldwide could die each year by 2050 from antibiotic resistant bacteria alone if no action is taken”, the United Nations Interagency Coordination Group on Antibiotic Resistance states in their report.
Antibiotics are miracle drugs.
It is a common misconception that antibiotics are the solution to all inflammatory conditions, not without reason. Inflammations caused by bacterial infection are treated so well with antibiotics. It seems only yesterday that whenever I had a bad cough or maybe even pharyngitis or bronchitis as a child, my granny would insist that I take some antibiotics. Yet, this kind of thinking might be contributing to a global crisis.
Ever since the discovery of penicillin in the last century, antibiotics have utterly transformed the field of medicine. Bacterial infections are no longer among the most life threatening conditions nor the highest cause of deaths.
Thanks to Fleming, the discoverer of penicillin, and all the science and research that came afterwards, nowadays most of us live long enough to die in other painful ways like cancer or Alzheimer’s rather than septic shock.
It is not so common in medicine to have a drug that solves a problem so straightforwardly and effectively. Yet, for severe bacterial infections and many other conditions, they are the only good solution.
In fact, antibiotics are required for an exceptionally wide spectrum of patients. From surgical patients to diabetic patients, from dialysis patients to even cancer patients, everyone has the possibility to gain infections at some point due to their preexisting conditions and vulnerable status.
Imagine having only weapon in your arsenal, let’s say a bow and some arrows, and they work like a charm.
According to a report by the CDC from 2018, the average length of stay for hospitalized patients with pneumonia (including the ICU) is just 7 days. Taking into account the fact that a large portion of these patients have viral pneumonias which take longer to treat, it is safe to say that a condition as serious as bacterial pneumonia can be treated in just a matter of days with the help of antibiotics.
The magic is fading.
Well then, what’s the problem? Picture that wooden arrow in your quiver that defeated so many enemies so easily, just bouncing right off them this time. You switch to your steel arrow and shoot again. The same thing happens. In desperation, you pull out the golden arrow that you never had to use, knowing that if it fails, you'll have nothing else that could work.
This is exactly what happens with antibiotic resistance. When the constantly mutating bacteria are exposed to the same medications over and over, they have a chance of developing defense mechanisms. Those bacteria can then either divide and spread or exchange genetic information with existing bacteria to “teach” resistance. In cases of extreme antibiotic resistance, doctors resort to last-line antibiotics to save the patient’s life.
Unfortunately, this still isn’t enough for some patients. It’s daunting not just for the patients but for doctors and public heath officials as well, witnessing the only effective treatment option fail on a condition so treatable. Nearly 3 million new infections in the United States each year are attributed to resistant strains of pathogens and an estimated 36,000 savable lives are lost. This is not just threatening to any single country. It is a global crisis.
Every 4 hours, the CDC detects a resistant germ that requires a public health investigation. The 2019 AR threat report shows the strains of bacteria requiring urgent attention and their rising numbers. In the worst cases, the percentage of resistance among a strain of bacteria to a certain antibiotic can increase by as much as 9% in only 2 years. Alternative methods such as antibody therapies work, yet require complex preparations and are too expensive to be readily available. Vaccines are only preventative and in some cases even discouraged, such as the BCG vaccine for tuberculosis, due to risk of potential side effects or influence in diagnostics of the actual disease.
Public health authorities around the world issue strict rules and guidelines on how antibiotic use should be approached. The drugs of last resort are never used as first-line treatment to decrease the likelihood of bacteria developing resistance to them. This of course makes a night and day difference. Nevertheless, as long as those antibiotics are being used, guidelines and precautions are only delaying the inevitable. New antibiotics are a constant necessity for this reason.
Pharmaceutical companies are racing against mutating bacteria to be one step ahead of antibiotic resistance. Or are they?
As a matter of fact, no new major classes of antibiotics have been discovered and introduced since 1987. This period is known as the “discovery void”.
All new drugs that came after were analogues, variants of drugs within the same classes.
On a related note, 78% of major drug companies have scaled back or completely abandoned their antibiotic developments since 1990, which explains the lack of new antibiotics. The substantial financial costs, scientific challenges and regulatory barriers combined with the relative low consumption of last-line drugs and less revenue isn’t the most appealing business model for pharmaceutical developers. Unfortunately, that would not be easily changed.
Artificial intelligence might just be the game changer here.
The technicality of machine learning algorithms are very complex, but the concept is easy to understand. Traditionally, a written program simply does what programmers tell it to do. If we want to sort out all numbers containing the number 5 from one to a million, we can write a program for it instead of doing it manually to save energy and time.
Machine learning algorithms are capable of doing much more than simply following instructions. By feeding them data sets, also known as training data, algorithms are able to analyze patterns and correlations from these data, learn how to interpret them by themselves. Trained models can later be used to make predictions or pick out similar entries that they think belong to the same categories from a much broader set of data, with greater efficiency and less cost. They don’t just do, they learn and they think.
Ever wondered how Google recommends images similar to the ones you are searching for? It uses algorithms that scan countless images to learn what makes two images similar or different. The more data they are fed, the more capable they become at finding images that are similar to what we’re already looking at. As far fetched as it may seem, finding a new antibiotic, in principle, isn’t that different.
Halicin: A win for medicine, a win for AI.
Thanks to a success story by some brilliant researchers at MIT Jameel Clinic, using AI to discover new drugs is no longer just some puffed up idea trying to cash in on hype.
According to their paper published in the Feburary 2020 issue of Cell , the team used a collection of 2,335 molecules that inhibit the growth of E.coli to train their model. Adding a few augmentations, they applied the model to several chemical libraries comprising of more than 107 million molecules.
Out of the numerous promising candidates that the algorithm spit out, one molecule stood out. An enzyme inhibitor, originally researched for the treatment of diabetes but discontinued due to poor results, proved to be a potent inhibitor of E.coli growth despite being very structurally different from conventional antibiotics.
Further investigations showed that this drug, later renamed halicin, proved to be a whole new class of antibiotic based on it’s unique, previously unknown mechanism of action. Astonishingly, the mechanism through which it killed bacteria was so competent that researchers could not reverse engineer it by studying halicin-resistant mutant bacteria, since their attempts to produce any failed.
Through hypotheses and experimentation, it was concluded that halicin possessed its antimicrobial activity by disrupting the electrochemical trans-membrane gradient that was crucial for bacterial survival. The drug tested potent for a broad spectrum of bacteria, including several multidrug-resistant strains the WHO considers most deadly. Clinical trials are already underway.
Revolution inbound.
Conventionally, candidates for new antimicrobial drugs are searched with a general category of chemicals in mind, usually where researchers expect to find new substances with already known mechanisms of action. Even just a couple thousand possibilities can take up to years to test, not to mention the cost. There is always the risk of not finding anything viable, in which case the whole process would need to be repeated again.
The discovery of halicin doesn’t just prove that artificial intelligence can help researchers screen a hundred times more molecules in a blink of an eye, at just a fraction of the cost. With the correct guidance and inputs, halicin is proof that deep learning algorithms are capable of searching in domains no one would ever think to look, making completely new discoveries that otherwise might have never been uncovered through conventional methods.
In fact, this revolutionary method doesn’t just apply to antibiotics. In Finland, machine learning models are being developed by researchers to predict drug combinations that kill cancer cells. It is only a matter of time before all kinds of drug research utilize artificial intelligence in some way or the other.
Revenue driven pharmaceutical companies would be strongly incentivized to make use of this technology and push it to its limits. AI can help find the best product while also shortening the development cycle and development costs. This might also provide a second chance for drugs previously deemed unprofitable and abandoned by Big Pharma. Either way, as long as they aren’t getting ripped off, patients will only benefit from the better drugs and increased options.
It is never not a joy witnessing different branches of science working together to create something spectacular. Certainly, there will be many more success stories to look forward to in the near future. I am optimistic that the biggest struggles of humanity today will become a thing of the past tomorrow, because when human beings work together, the possibilities are limitless.
References
- Williams S, Gousen S, DeFrances C. National Hospital Care Survey Demonstration Projects: Pneumonia Inpatient Hospitalizations and Emergency Department Visits. Natl Health Stat Report. 2018;(116):1–11.
- New report calls for urgent action to avert antimicrobial resistance crisis. [online] Who.int. Available at: <https://www.who.int/news/item/29-04-2019-new-report-calls-for-urgent-action-to-avert-antimicrobial-resistance-crisis> [Accessed 28 January 2022].
- CDC. Antibiotic Resistance Threats in the United States, 2019. Atlanta, GA: U.S. Department of Health and Human Services, CDC; 2019
- Coates AR, Halls G, Hu Y. Novel classes of antibiotics or more of the same?. Br J Pharmacol. 2011;163(1):184–194. doi:10.1111/j.1476–5381.2011.01250.x
- Cooper, M.A. and D. Shlaes, Fix the antibiotics pipeline. Nature, 2011. 472(7341): p. 32–32.
- Stokes JM, Yang K, Swanson K, et al. A Deep Learning Approach to Antibiotic Discovery [published correction appears in Cell. 2020 Apr 16;181(2):475–483]. Cell. 2020;180(4):688–702.e13. doi:10.1016/j.cell.2020.01.021