Imagine you are a fossil hunter. You spend months in the heat of Arizona digging up bones just to discover that what you’ve discovered is from a previously discovered dinosaur.
This is how the search for antibiotics has recently turned out. The relatively few antibiotic hunters continue to find the same types of antibiotics.
With the rapid increase in drug resistance in many pathogens, new antibiotics are desperately needed. It may only be a matter of time before an injury or scratch becomes a threat to life.
However, few new antibiotics have entered the market in recent times, and even these are only minor variants of old antibiotics.
While the outlook seems bleak, the recent revolution in artificial intelligence (AI) offers new hope. In a study published on Feb. 20 in the journal Cell, MIT and Harvard scientists used a type of AI called deep learning to discover new antibiotics.
The traditional way of discovering antibiotics, from extracts of soil or plants, has not revealed new candidates, and there are also many social and economic obstacles to solve this problem.
Some scientists have tried to address it recently by looking for new antibiotic-producing genes in the DNA of bacteria. Others look for antibiotics in exotic places like in our noses.
Drugs found through unconventional methods face a difficult path to reach the market. Medications that are effective in a Petri dish may not work well inside the body.
They may not be well absorbed or have side effects. The manufacture of these medicines in large quantities is also an important challenge.
Enter deep learning. These algorithms enhance many of the current facial recognition systems and autonomous cars. Imitate how neurons operate in our brain by learning patterns in the data.
An individual artificial neuron, such as a mini sensor, can detect simple patterns such as lines or circles. By using thousands of these artificial neurons, deep learning artificial intelligence can perform extremely complex tasks, such as recognizing cats in videos or detecting tumors in biopsy images.
Given its power and success, it may not be surprising to know that researchers looking for new medications are adopting artificial intelligence of deep learning. However, building an AI method to discover new medications is not a trivial task. In large part, this is because there is no free lunch in the AI field.
The No Free Lunch theorem states that there is no universally superior algorithm. This means that if an algorithm works spectacularly in one task, for example, facial recognition, it will spectacularly fail in a different task, such as drug discovery. Therefore, researchers cannot simply use standard deep learning AI.
The Harvard-MIT team used a new type of deep learning AI called neural networks of graphs for drug discovery. In the stone age of AI in 2010, AI models for drug discovery were constructed using text descriptions of chemicals. This is like describing a person’s face through words like “dark eyes” and “long nose.”
These text descriptors are useful, but they obviously do not paint the entire image. The AI method used by the Harvard-MIT team describes the chemicals as a network of atoms, which gives the algorithm a more complete picture of the chemical than text descriptions can provide.
Human knowledge and AI blank whiteboards
However, deep learning alone is not enough to discover new antibiotics. It must be combined with a deep biological knowledge of infections.
The Harvard-MIT team meticulously trained the AI algorithm with examples of drugs that are effective and those that are not. In addition, they used drugs that are known to be safe in humans to train AI.
They then used the AI algorithm to identify potentially safe but potent antibiotics of millions of chemicals.
Unlike people, AI has no preconceived notions, especially about how an antibiotic should be. Using old-school AI, my laboratory recently discovered some surprising candidates for the treatment of tuberculosis, including an antipsychotic medication.
In the Harvard-MIT team study, they found a gold mine of new candidates. These candidate medications are nothing like existing antibiotics. A promising candidate is Halicin, a medication that is being explored to treat diabetes.
Halicin, surprisingly, was potent not only against E. coli, the bacteria in which the AI algorithm was trained, but also in more deadly pathogens, including those that cause tuberculosis and inflammation of the colon.
In particular, halicin was potent against resistant medications. Acinetobacter baumanni. This bacterium tops the list of deadliest pathogens compiled by the Centers for Disease Control and Prevention.
Unfortunately, Halicin’s broad potency suggests that it can also destroy harmless bacteria in our body. It can also have metabolic side effects, as it was originally designed as an antidiabetic medication. Given the extreme need for new antibiotics, these can be small sacrifices to pay to save lives.
Stay ahead of evolution.
Given Halicin’s promise, should we stop the search for new antibiotics?
Halicin could eliminate all obstacles and eventually reach the market. But he still needs to overcome an implacable enemy that is the main cause of the drug resistance crisis: evolution.
Humans have thrown numerous drugs at the pathogens during the last century. However, pathogens have always developed resistance. Therefore, it is likely that it will not take long until we find a halicin-resistant infection.
However, with the power of deep learning AI, we can now be better prepared to respond quickly with a new antibiotic.
There are many challenges ahead for possible antibiotics discovered using AI to reach the clinic. The conditions under which these medications are tested are different from those inside the human body.
My laboratory and others are building new AI tools to simulate the internal environment of the body and evaluate the potency of antibiotics. AI models can also predict the toxicity of medications and side effects.
These artificial intelligence technologies together will soon give us an advantage in the never-ending battle against drug resistance.
[[[[You are intelligent and curious about the world. So are the authors and editors of The Conversation. You can read us daily by subscribing to our newsletter.]
Sriram Chandrasekaran, Assistant Professor of Biomedical Engineering, University of Michigan.
The Conversation republishes this article under a Creative Commons license. Read the original article.