On Tuesday, the Nobel Committee for Physics honoured two physicists, John Hopfield and Geoffrey Hinton, for laying the foundations of Artificial Intelligence (AIF) through the modelling of Artificial Neural Network (ANN). On Wednesday, the Nobel Committee for Chemistry chose David Baker at the University of Washington in Seattle, the United States, and Demi Hassabis and John Jumper at Deep Mind at Google, London, Britain, the 2024 Nobel Prize for Chemistry for creating the Artificial Intelligence (AI) software, called Rosetta by Baker, and AlphaFold2 by Hassabis and Jumper to predict the protein structure of all the possible proteins, 200 million so far, of all organisms, including human.
While Baker’s Rosetta is based on the protein patterns, AlphaFold2 creates three-dimensional models of all the proteins. Baker incorporated the AlphaFold 2 into his Rosetta called RoseTTAFold to improve the prediction of protein structures. The methods of Baker, Hassibis and Jumper speed up the discovery of new proteins, which will help in faster drug development. But there is a caveat.
While in the majority of cases, AlphaFold2 has turned out to be accurate or accurate enough, it has been found that the AI product cannot predict if there is no structure of the protein. This happens quite often in the human organism. The deformed protein leads to diseases like cancer. The AI mode of predicting protein structures has to be backed by the older methods of X-ray crystallography and electron microscopy testing real-world protein structures. There is this fact that AlphaFold2 has run through the test of Critical Assessment of Protein Structure Prediction (CASP), and it has won the competition in 2020, and the beginning of the “Digital Age of Biology” was proclaimed.
An interesting aspect of the Nobel Prize for Chemistry this year is that the AI products, Rosetta and AlphaFold2, have been of recent making. While AlphaFold had debuted in 2018, Baker had come up with Rosetta a decade earlier. And improvements are being made in both the AI products. Google has also created a free AlphaFold database of the protein structures which researchers can access. The availability of millions of protein structures in their virtual form makes things easy for the researchers in the universities as well as in the research and development departments of the pharmaceutical companies.
But there is need for caution. The virtual models of proteins generated by AlphaFold and RoseTTAFold are accurate but their properties and how they would interact with other combinations of proteins remains an uncharted territory.
Baker, Hassibis and Jumper came from unorthodox academic backgrounds and interests. Baker started off with philosophy and social science. After a course in evolutionary biology, he became interested in cell biology and then was fascinated by protein structures. He started with chess and became a master at the age of 13. He became a programmer and games developer. Then he got interested in neuroscience, and he used this knowledge to develop the artificial neural network (ANN). Jumper was interested in the universe and started with physics and mathematics. Then he joined a company where supercomputers were used to “simulate proteins and their dynamics”. He kept his interest in proteins when he did his doctorate in theoretical physics.
There are many things at work here. The traditional divisions between the scientific disciplines are breaking down. The American higher education system seems to keep the doors open from all sides. And the computers, and from the computers the AI, have become creative tools in probing scientific secrets. It does appear that AI is the magic key that will open all doors of knowledge. It is an exciting prospect indeed.