Predicting the structure of proteins is one of the tasks of research in the field of biology. The MIT Technology Review published an announcement by Deepmind about its artificial intelligence tool, the AlphaFold program. The company claims to have successfully predicted the structure of almost every protein known to scientists and is offering free access to its database of more than 200 million proteins to anyone. At the same time, MIT researchers published a study, “A Comparative Analysis of AlphaFold-Enabled Molecular Docking Predictions for Antibiotic Discovery,” highlighting that improvements will be needed to fully utilize the protein structures provided by AlphaFold.
Deepmind offered the 1st version of AlphaFold, an AI system for predicting protein structures on CASP (Critical Assessment of protein Structure Prediction) in 2018, it took 1st place, as did AlphaFold 2, the second version of which we also devoted an article in our paper ActuIA N°3 magazine, did it in 2020.
According to Deepmind, AlphaFold, its artificial intelligence system that predicts the three-dimensional structure of a protein based on its amino acid sequence, consistently achieves competitive accuracy compared to experiment.
In 2021, she published a scientific paper and source code explaining how she created this artificial intelligence system, and partnered with the European Bioinformatics Institute EMBL (EMBL-EBI) to create the AlphaFold platform database to make these predictions free to the scientific community. The latest version of the database contains over 200 million records, providing a wide coverage of UniProt, a standard repository of protein sequences and annotations.
The AlphaFold database is currently focused on the use case approved in CASP14: predicting the structure of a single protein chain with a natural sequence. Deepmind and EMBL will continue to update this database.
However, Deepmind acknowledges that the AlphaFold system has some limitations. MIT researchers sought to analyze them and their potential.
Study: Comparative Analysis of AlphaFold Docking Predictions for Antibiotic Discovery
A team of researchers led by James Collins, Thermeer Professor of Medical Engineering and Science at the Institute of Medical Engineering and Science (IMES) and MIT’s Department of Biological Engineering, investigated whether AlphaFold could predict interactions between bacterial proteins and antibacterial properties. compounds that could enable the development of new antibiotics.
Efficient identification of drug mechanisms of action remains a challenge. Computational docking approaches are widely used to predict drug binding targets. However, such approaches depend on existing protein structures, and accurate structural predictions are only available in AlphaFold2.
For this study, the researchers combined AlphaFold2 with molecular docking modeling to predict protein-ligand interactions between 296 proteins spanning the Escherichia coli core proteome and 218 active antibacterial compounds and 100 inactive compounds, respectively, indicating promiscuous compounds and proteins.
They then compared the performance of the model by measuring the enzymatic activity of 12 major proteins treated with each antibacterial compound.
This allowed them to confirm extensive promiscuity, but also found an average area under the receiver operating characteristic curve (auROC) of 0.48, indicating poor model performance. The researchers demonstrated that re-estimating berthing positions using machine learning-based approaches improves model performance, resulting in an average auROC of 0.63. They also found that the correction feature sets improved the accuracy of the prediction and the true positive to false positive ratio.
A) To determine the chemical space of interest, the researchers conducted high-throughput growth inhibition screens for wild-type E. coli. Growth inhibitory compounds were considered active, and each active compound was docked to each of the 296 major E. coli protein structures predicted by AlphaFold2. B) Growth inhibition measurements for 39,128 compounds, of which 218 compounds (including known antibiotics) were identified as active against E. coli BW25113. Data are taken from two biological replicates. Compounds with an average relative growth of less than 0.2 were classified as active (red dots) and all other compounds were classified as inactive (blue dots). C) Distribution of compound classes represented by 218 active compounds.
The study demonstrated the potential of AI to streamline the selection process for future antibiotics. However, the results indicate that while AlphaFold2 can provide rich structural information, methods are needed to more accurately model protein-ligand interactions in order to better use AlphaFold2 for drug discovery.
Links to articles:
“Comparison of molecular docking predictions with AlphaFold support for antibiotic discovery” doi.org/10.15252/msb.202211081
Publication in the journal “Molecular Systems Biology” dated September 6, 2022
Wong F., Krishnan A., Zheng E.J., Stark H., Manson A.L., Earl A.M., Jaakkola T., Collins J.J. 18:e11081.
Felix Wong1,2,3, Artie Krishnan1,2,3, Erika J. Zheng3,4, Hannes Stark5, Abigail L. Manson3, Ashley M. Earl3, Tommy Jaakkola5 and James J. Collins 1,2,3,6
1Institute of Medical Engineering and Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
2Department of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
3Infectious Diseases and Microbiome Program, Broad Institute, Massachusetts Institute of Technology and Harvard, Cambridge, Massachusetts, USA
4Chemical Biology Program, Harvard University, Cambridge, Massachusetts, USA
5Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
6 Wiss Institute of Biological Engineering, Harvard University, Boston, Massachusetts, USA.