Current methods for screening potential drugs are too time consuming and expensive to develop new antibiotics. Computational models can be used to identify new drugs at a cheaper rate.
The MIT study shows the potential and limitations of a particular approach. The researchers looked at whether existing models could accurately predict the interactions between the two types ofbacteria. This type of modeling could be used to do large-scale screens for new compounds that target previously untargeted genes. The development of antibiotics with unprecedented mechanisms of action is a task essential to addressing the antibiotic resistant crisis.
The researchers, led by James Collins, the Termeer Professor of Medical Engineering and Science in MIT's Institute for Medical Engineering and Science, found that the existing models did not perform well for this purpose. They did not perform better than chance.
"AlphaFold is expanding the possibilities for in silico drug discovery efforts, but these developments need to be coupled with additional advances in other aspects of modeling that are part of drug discovery efforts." The current abilities and limitations of computational platforms for drug discovery were discussed in the study.
The researchers were able to improve the performance of the models by using machine- learning techniques. The researchers say that more improvement is needed to take full advantage of AlphaFold.
The study was written by Collins and was published in the journal. The paper is written by Felix Wong and Aarti Krishnan.
There are some interactions that are referred to as "molecular interactions".
The goal of the Antibiotics-AI Project is to use artificial intelligence to find and design new antibiotics.
AlphaFold, an artificial intelligence software developed by DeepMind and Google, has been able to predict the structure of some of the world's most important scientific discoveries. The AlphaFold structures can be used to find new antibiotics that bind to certain types ofbacteria.
Collins and his students decided to study the interactions of E coli with 218 antibacterial compounds, including antibiotics.
The researchers looked at how these compounds interact with E. coli.
This type of simulation can be used to screen compounds against a single target to find the best ones. The predictions turned out to be less accurate when the researchers were trying to screen many compounds against many potential targets.
The model had false positive rates similar to true positive rates when compared to interactions for 12 essential proteins. The model was not able to consistently identify true interactions between drugs and their targets.
Poor performance was found when using a measurement to evaluate models. Collins says that using the standard docking simulations, they got an AuROC value of roughly 0.5, which basically says you're doing no better than if you were randomly guessing.
The researchers found the same results when they used this approach with structures that have been determined.
If we're going to use AlphaFold effectively and extensively in drug discovery, we need to do a better job with docking models.
Predicting better.
One possible reason for the model's poor performance is that the model's structures are static, while in biological systems the structures are flexible.
The researchers ran the predictions through four more machine-learning models in order to improve the success rate. The data that is used to train these models allows them to incorporate more information into their predictions.
The docking predictions are used to assess the chemical and physical properties of the known interactions by the machine- learning models. The higher the ratio of true positives to false positives, the better.
More improvement is needed before this type of modeling can be used to identify new drugs. Training the models on more data would be one way to do this.
With further advances, scientists may be able to harness the power of artificial intelligence to find new antibiotics and other drugs to treat a variety of diseases. Improvements to the modeling approaches and expansion of computing power will make these techniques more important in drug discovery. We have a long way to go to achieve the full potential of in silico.
More information: Benchmarking AlphaFold-enabled molecular docking predictions for antibiotic discovery, Molecular Systems Biology (2022). DOI: 10.15252/msb.202211081 Journal information: Molecular Systems Biology
All Rights Reserved © 2024
