Artificial intelligence is changing the rules of structural biology science tech
“I never thought we would reach this point in my life.” Those are the words of a research leader in structural biology responding to research published last week that uses artificial intelligence to predict the structure of more than 20,000 human proteins, as well as nearly all known proteins produced by 20 model organisms, such as Escherichia coli and flies. Fruit and yeast, as well as soybeans and Asian rice. This is a grand total of about 365 thousand predictions.
Biology’s giant leap on the back of artificial intelligence
The data was first released to the public online on July 22 by a joint team of researchers at DeepMind, a London-based artificial intelligence company owned by parent company Google, Alphabet, and the European Institute of Bioinformatics, based in the Molecular Biology Laboratory. European near Cambridge, UK.
The DeepMind team has developed a machine learning tool called AlphaFold . The team trained this software on DNA sequences, including its evolutionary history, and the already known shapes of tens of thousands of proteins contained in a public access database of proteins hosted by European Molecular Biology Laboratory researchers.
A week ago, DeepMind also released the source code for AlphaFold and detailed how it was built , at the same time researchers from the University of Washington, Seattle published details of another protein structure prediction program inspired by AlphaFold called RoseTTAFold3 .
Revealing this catalog of predicted structures would not have been good news had the data and methodology not been open and freely available, as structural biologists and other researchers have already begun using AlphaFold to obtain more accurate models of proteins that are difficult or impossible to characterize with current experimental methods.
Read also: What is the Apple Pay service ?
Structure prediction acceleration
Predicting the 3D shape in which proteins bend has been one of the unresolved “big challenges” in biology since the discovery of the structure of DNA itself in 1953. Before artificial intelligence, predicting the structure from sequence was a time-consuming process, with little guarantee of obtaining Accurate result.
Although the new data needs to be validated and verified empirically, AI tools can accurately predict protein structures within minutes to hours, compared to months or years that traditional methods would take to determine the structure of just one or two proteins. It opens up huge potential for applications, for example in engineering enzymes to break down environmental pollutants such as microplastics.
Recent advances depend not only on sharing open data, but on advances in basic science and technology. Since the 1960s, structural biologists have worked on parallel approaches to understanding the science of protein folding, one of which involves assembling the structures of proteins through an understanding of fundamental physical forces.
There have been other attempts to predict shapes by making comparisons with closely related proteins using an organism’s evolutionary history, hence the important role of imaging techniques, from X-ray crystallography to cryo-electron microscopy.
In the basic sciences of structural biology, there are still major problems to be solved. Although AI in science and technology is good at showing accurate results, it does not explain (at least for the time being) how or why these results occurred.
The teams at DeepMind, the European Molecular Biology Laboratory, the University of Washington and other participating institutions must be congratulated on the crucial achievements, but there is still work to be done to unlock the information about how and causality of proteins bend.
Public access and end of monopoly
In terms of importance, some compare these recent advances to the first draft of the human genome sequence 20 years ago, and it is true that there are comparisons to be made between the two achievements. Both the Human Genome Project and the DeepMind Catalog of Human Protein Structure Predictions are equipping their fields with a tool set to significantly accelerate discovery.
The first draft of the human genome was the result of a race, and the protein folding solution also benefited from a kind of competition, through an annual event called Critical Assessment of Structure Prediction of Protein (CASP), which was necessary to get the latter result.
Research teams today, just like those involved in early genome sequencing, needed open access to data. By making the data and methodology available to everyone, DeepMind is now setting a standard that will make it difficult for other companies in the field, such as Facebook and Microsoft, to continue arguing over the monopoly data.
So, what about the future? Over the past week, the scientific journal Nature has interviewed nearly a dozen researchers in the field, most of whom agree that it is too early to predict exactly what impact the application of AI will have in the life sciences, except that whatever the impact, it will be absolutely transformative. .
Accurately predicting how AI will change biology needs good training data, which we don’t yet have, but in AI, the structural biology research community—and its collaborators in other fields—have a wealth of new data.
In addition to its research and data, AI provides a window into models for organizing and managing research that universities should study, and for today’s researchers and future generations, there is a lot of work to follow.
Artificial intelligence is changing the rules of structural biology science tech
Artificial intelligence is changing the rules of structural biology science tech
