Analyze AlphaFold's potential for drug discovery

Few new antibiotics have been developed over the past few decades, largely because the current methods of screening potential drugs are too expensive and time-consuming

A new study from the Massachusetts Institute of Technology reveals the potential and limitations

However, researchers led by James Collins, a professor at MIT's Institute of Medical Engineering and Sciences (IMES) and the Department of Bioengineering, found that these existing models do not perform well

"Breakthroughs like AlphaFold are expanding the possibilities of silicon drug discovery efforts, but these developments need to be combined with additional advances in other aspects of modeling drug discovery efforts," Collins said

In their new study, the researchers were able to improve the results by applying machine learning techniques to improve the performance

Collins is the senior author of the study, which was published today in the journal Molecular Systems Biology

Molecular interactions

The new research is part of the Antibiotic-ARTIFICIAL Intelligence Project, recently launched by Collins Labs, which aims to use AI to discover and design new antibiotics

AlphaFold, an artificial intelligence software jointly developed by DeepMind and Google, has accurately predicted protein structure from amino acid sequences

To test the feasibility of this strategy, Collins and his students decided to study the interaction

The researchers analyzed the interaction of these compounds with E.

This simulation has been successfully used in the study of screening a large number of compounds against a single protein target to determine the compound that binds

By comparing the model's predictions with the actual interactions of 12 essential proteins obtained from laboratory experiments, the researchers found that the model's false-positive rate was similar to

Through auROC, a method commonly used to evaluate computational models, the researchers also found poorer performance

When the researchers used this modeling method on the experimentally determined protein structure, they found similar results, rather than the structure predicted by

"AlphaFold seems to have roughly the same structure as the experiments have determined, but if we're going to make effective and extensive use of AlphaFold in drug discovery, we need to do a better job with molecular docking models," Collins said

Better predictions

One possible reason for poor model performance is that the protein structure of the input model is static, while in biological systems, proteins are flexible and often change their configuration

To improve the success rate of the modeling method, the researchers made predictions
using four additional machine learning models.
These models are based on data that describes how proteins and other molecules interact, enabling them to incorporate more information into the predictions
.

"Machine learning models learn not only shapes, but also chemical and physical properties of known interactions, and then use that information to re-evaluate docking predictions," Wong said
.
"We found that if you use these additional models to filter out interactions, you can get a higher ratio of
true positives and false positives.
"

However, the researchers say that further improvements
are needed before this type of model can be used to successfully identify new drugs.
One approach is to train the model with more data, including the biophysical and biochemical properties of the proteins, as well as their different conformations, and how these features affect their binding
to potential drug compounds.

As further advances, Collins said, scientists may be able to harness the power of the protein structures produced by AI to discover not only new antibiotics but also drugs
to treat a variety of diseases, including cancer.
"We are optimistic that these technologies will become increasingly important
in drug discovery as modeling methods improve and computing power expands," he said.
"However, we still have a long way to go
to realize the full potential of silicon drug discovery.
"

essay

Benchmarking AlphaFold-enabled molecular docking predictions for antibiotic discovery