Tuesday, September 27

Analyzing AlphaFold’s Potential in Drug Discovery | MIT News

In recent decades, very few new antibiotics have been developed, largely because current methods of screening for potential drugs are extremely expensive and time-consuming. A promising new strategy is to use computer models, which offer a potentially faster and cheaper way to identify new drugs.

A new study from MIT reveals the potential and limitations of such a computational approach. Using protein structures generated by an artificial intelligence program called AlphaFold, researchers explored whether existing models could accurately predict interactions between bacterial proteins and antibacterial compounds. If so, then researchers could begin to use this kind of modeling to perform large-scale screens for new compounds targeting previously untargeted proteins. This would allow the development of antibiotics with unprecedented mechanisms of action, a key task in tackling the antibiotic resistance crisis.

However, the researchers, led by James Collins, Termeer Professor of Medical Engineering and Science at the Institute of Medical Engineering and Sciences (IMES) and the Department of Biological Engineering at MIT, found that these existing models do not did not work well for this purpose. In fact, their predictions worked out little better than chance.

“Breakthroughs such as AlphaFold expand the possibilities for in silico drug discovery efforts, but these developments must be coupled with additional advances in other aspects of modeling that are part of drug discovery efforts,” says Collins. “Our study speaks to both the current capabilities and the current limitations of computational platforms for drug discovery.”

In their new study, the researchers were able to improve the performance of these types of models, known as molecular docking simulations, by applying machine learning techniques to refine the results. However, further improvements will be needed to take full advantage of the protein structures provided by AlphaFold, according to the researchers.

Collins is the lead author of the study, which appears today in the journal Biology of molecular systems. MIT postdocs Felix Wong and Aarti Krishnan are the lead authors of the paper.

Molecular interactions

The new study is part of a recently launched effort by Collins’ lab called the Antibiotics-AI Project, which aims to use artificial intelligence to discover and design new antibiotics.

AlphaFold, an AI software developed by DeepMind and Google, accurately predicted the structures of proteins from their amino acid sequences. This technology has sparked excitement among researchers looking for new antibiotics, who hope to use AlphaFold structures to find drugs that bind to specific bacterial proteins.

To test the feasibility of this strategy, Collins and his students decided to study the interactions of 296 essential proteins of E.coli with 218 antibacterial compounds, including antibiotics such as tetracyclines.

The researchers analyzed how these compounds interact with E.coli proteins using molecular docking simulations, which predict the bond strength of two molecules based on their shapes and physical properties.

This type of simulation has been used successfully in studies that screen a large number of compounds against a single protein target, to identify compounds that bind best. But in this case, where the researchers were trying to screen many compounds against many potential targets, the predictions turned out to be much less accurate.

By comparing predictions produced by the model with actual interactions for 12 key proteins, obtained from laboratory experiments, the researchers found that the model had similar false positive rates to true positive rates. This suggests that the model was unable to consistently identify the true interactions between existing drugs and their targets.

Using a metric often used to evaluate computer models, known as auROC, the researchers also found poor performance. “Using these standard molecular docking simulations, we got an auROC value of around 0.5, which basically means you’re not doing any better than guessing at random,” Collins says.

The researchers found similar results when they used this modeling approach with experimentally determined protein structures, instead of the structures predicted by AlphaFold.

“AlphaFold appears to do about as well as experimentally determined structures, but we need to do a better job with molecular docking models if we are to use AlphaFold effectively and widely in drug discovery,” Collins says.

Better predictions

One of the possible reasons for the poor performance of the model is that the protein structures introduced into the model are static, whereas in biological systems proteins are flexible and often change configuration.

To try to improve the success rate of their modeling approach, the researchers ran the predictions on four additional machine learning models. These models are trained on data that describes how proteins and other molecules interact with each other, allowing them to incorporate more information into predictions.

“Machine learning models not only learn the shapes, but also the chemical and physical properties of known interactions, then use that information to reassess docking predictions,” Wong says. “We found that if you were to filter interactions using these additional patterns, you can get a higher ratio of true positives to false positives.”

However, further improvements are still needed before this type of modeling can be used to successfully identify new drugs, the researchers say. One way to do this would be to train the models on more data, including the biophysical and biochemical properties of proteins and their different conformations, and how these characteristics influence their binding to potential drug compounds.

This study both allows us to understand how far we are from realizing complete machine learning-based paradigms for drug development, and provides fantastic experimental and computational benchmarks to stimulate, direct and guide progress towards this future vision,” says Roy Kishony, professor. in biology and computer science at the Technion (the Israel Institute of Technology), which was not involved in the study.

With new advances, scientists may be able to harness the power of AI-generated protein structures to discover not only new antibiotics, but also drugs to treat various diseases, including cancer, Collins said. “We are optimistic that with improvements in modeling approaches and expansion of computing power, these techniques will become increasingly important in drug discovery,” he says. “However, we still have a long way to go to reach the full potential of in silico drug discovery.”

The research was funded by the James S. McDonnell Foundation, the Swiss National Science Foundation, the National Institute of Allergy and Infectious Diseases, the National Institutes of Health, and the Broad Institute of MIT and Harvard. The Antibiotics-AI project is supported by the Audacious project, the Flu Lab, the Sea Grape Foundation and the Wyss Foundation.