"The scientific Nobel prizes have always, in their way, honoured human intelligence. This year, for the first time, the transformative potential of artificial intelligence (AI) has been recognised as well. That recognition began on Tuesday October 8th, when Sweden’s Royal Academy of Science awarded the physics prize to John Hopfield of Princeton University and Geoffrey Hinton of the University of Toronto for computer-science breakthroughs integral to the development of many of today’s most powerful AI models.
The next day, the developers of one such model also received the coveted call from Stockholm. Demis Hassabis and John Jumper from DeepMind, Google’s AI company, received one half of the chemistry prize for their development of AlphaFold, a program capable of predicting three-dimensional protein structure, a long-standing grand challenge in biochemistry. The prize’s other half went to David Baker, a biochemist at the University of Washington, for his computer-aided work designing new proteins.
The AI focus was not the only thing the announcements had in common. In both cases, the research being awarded would be seen by a stickler as being outside the remit of the prize-giving committees (AI research is computer science; protein research arguably counts as biology).
Boundary-pushing
Such flexibility is not unprecedented. In 1973, for example, three pioneering students of animal behaviour, who worked on honeybees, geese and sticklebacks, were shoehorned into the physiology category. The award to Drs Hinton and Hopfield, however, recognises achievements with more profound consequences.
Both researchers did their crucial work in the early 1980s, at a time when computer hardware was unable to take full advantage of it. Dr Hopfield was responsible for what has become known as the Hopfield network—a type of artificial neural network that behaves like a physical structure called a spin glass, which gave the academy a tenuous reason to call the field “physics”. Dr Hinton’s contribution was to use an algorithm known as backpropagation to train neural networks.
Artificial neural networks are computer programs based loosely on the way in which real, biological networks of nerve cells or neurons are believed to work. In particular, the strengths of the connections (known as weights) between “nodes” (the equivalent of neurons) in such networks are plastic. This plasticity grants a network the ability to process information differently in response to past performance; or, in other words, to learn. Hopfield networks, in which each node is connected to every other except itself, are particularly good at learning to extract patterns from sparse or noisy data.
Dr Hinton’s algorithm turbocharged neural networks’ learning ability by letting them work in three dimensions. Hopfield networks and their ilk are, in essence, two-dimensional. Though they actually exist only as simulations in software, they can be thought of as physical layers of nodes. Stack such layers on top of one another, though, and train them by tweaking the weights as signals move both backward and forward between the layers (ie, back-propagated as well as forward-propagated) and you have a much more sophisticated learning system.
Dr Hinton also, for good measure, tweaked Dr Hopfield’s networks using a branch of maths called statistical mechanics to create what are known as Boltzmann machines. (Statistical mechanics, which underlies modern understanding of the second law of thermodynamics, was invented by Ludwig Boltzmann, a near contemporary of Alfred Nobel.) Boltzmann machines can be used to create systems that learn in an unsupervised manner, spotting patterns in data without having to be explicitly taught.
It is, then, the activities of these two researchers which have made machine learning really sing. AI models can now not only learn, but create (or, for sceptics, reorganise and regurgitate in a most sophisticated manner). Such tools have thus gone from being able to perform highly specific tasks, such as recognising cancerous cells in pictures of tissue samples or streamlining mountains of particle-physics data, to anything from writing essays for lazy undergraduates to running robots.
Dr Hinton, whom the academy’s detectives tracked down to a hotel in California to deliver the glad tidings, and who gamely agreed to answer questions from the press, despite the time difference, seemed both worried and proud about his achievements. He fretted, as many in the field do, about how machine intelligence that outstripped the human variety would then go on to treat its creators. But he also mused that by assisting mental labour, AI might have as big an effect on society as the Industrial Revolution’s assistance of physical labour has had.
Such musings were timely. Not 24 hours later, the academy would recognise research conducted, with the help of AI models, on the structure of proteins.
Return to the fold
Proteins are the main chemical building blocks of life. They are made up of smaller molecules called amino acids, arranged in long chains which fold in highly complex and specific ways. A protein’s final folded form determines its biological function. In other words, to understand proteins—and, by extension, biology—one must understand their structure.
Dr Baker achieved such understanding through doing. In a landmark paper from 2003, he succeeded in designing a completely new protein. Using a computer program he had named Rosetta, he found an amino-acid sequence capable of folding in ways not seen in nature. Once the sequence was recreated in the lab and the protein formed, he determined its final structure using a technique called X-ray crystallography: it was a close match to what he had set out to make. Rosetta, now called Rosetta Commons, has subsequently become a software package used by protein chemists around the world, and computational protein design has assisted in everything from vaccine development to the detection of toxic chemicals.
Going the other way, and predicting a protein’s structure from its amino-acid sequence, is a problem that took even longer to crack. Given the near-limitless number of configurations into which a protein can fold—by some estimates, as many as 10300 for a single complex protein—even computers had limited success. DeepMind’s AlphaFold 1 and 2 (both artificial neural networks), made public in 2018 and 2020 respectively, were the first to even get close. AlphaFold 2 now has a database of more than 200m protein structure predictions, with a prediction accuracy approaching 90%.
Though Sir Demis and Dr Jumper have featured on various contender lists this year, many wondered if it was too soon for AlphaFold to be recognised. Yet it has already had real impact: DeepMind says that some 2m scientists already use it in their research. AlphaFold 3, released in May, goes beyond proteins to predict the structure of a host of other biomolecules, such as DNA, as well as small molecules that might function as drugs. It can also predict how different molecules with different structures fit together, such as how a virus’s spike protein might interact with antibodies and sugars found in the body.
By choosing, for the first time, to honour work performed with an AI model, the committee has opened the door for more such prizes in the future. That is just as well; AI has been seeping into all areas of science for some time now, as Dr Baker illustrated when he was phoned up during the committee’s press conference. He said that AlphaFold has inspired him to make generative-AI models that can design new proteins. “Our new AI methods are much more powerful,” he said.
Mega-important
The prize for physiology or medicine, for its part, eschewed any mention of AI. It was also immune to charges of genre-bending, continuing the academy’s trend of increasingly recognising the smallest advances at the molecular and cellular level—rather than work on physiology or organs—because it is on these microscopic scales that the most exciting scientific frontiers are to be found.
The joint winners were Victor Ambros at the University of Massachusetts Medical School and Gary Ruvkun at Massachusetts General Hospital for their discovery of micro-RNA (miRNA) and its role in “post-transcriptional gene regulation”. These are a class of small molecules composed of only 20 to 24 nucleotides (the A, C, G, U letters of the genome), and they play a key role in how cells work.
Inside the nucleus of every human cell is a full set of instructions—the genome—for creating a person. A key question in biology is how the same set of genes and instructions can lead to such different types of cells in the body, from muscle to liver cells by way of the neurons found in the brain. The answer is that not all the genes within a nucleus are translated into protein. Different types of cells follow their own developmental pathways by using only those genetic instructions relevant to their growth and development. The selection necessary for each cell type is controlled in part by the miRNA molecules discovered by Drs Ambros and Ruvkun.
They work primarily by binding to target parts of another molecule within cells, known as messenger RNA (mRNA)—which carries information from the DNA of the genome to the protein-making factories within cells. By interfering with mRNA molecules, miRNA can alter or prevent the production of proteins. Underscoring the growing importance of this area of molecular biology, mRNA was itself the subject of the Nobel prize last year.
Finding miRNAs, in 1993, paved the way to the understanding, today, that there are over a thousand of these small molecules within our cells. The discovery has had far-reaching implications in biology. Abnormal regulation by miRNA molecules can contribute to cancer and epilepsy. Mutations in genes that code for miRNA molecules have been found to cause conditions such as congenital hearing loss and are thought to be involved in the pathology of many eye disorders, such as cataracts, glaucoma and macular degeneration. The miRNA molecules are also thought to be important in numerous bone diseases, such as osteoporosis, osteosarcoma and bone metastasis.
Drs Ambros and Ruvkun—who worked at the same lab in the late 1980s at the Massachusetts Institute of Technology—discovered miRNA molecules using a key tool of biological inquiry: the roundworm Caenorhabditis elegans. They were studying two mutant strains of worms that had defects in the genes that dictated how the animals developed and worked. In doing so, the researchers showed that a gene called lin-4 produced an unusually short RNA molecule that did not code for any proteins and which seemed to inhibit the activity of another gene.
In awarding the prize, the Karolinska Institute’s Nobel committee noted that when the scientists published their results, they met an “almost deafening silence from the scientific community”. The unusual mechanism of gene regulation in C. elegans was assumed to be a peculiarity of that organism, not relevant to humans or other more complex organisms. That view eventually shifted, as it became clear that genes that encode for miRNAs were found throughout the animal kingdom.
Novo Nordisk, a Danish pharmaceutical giant, is one of the firms trying to make medicines using miRNAs. This year it acquired Cardior, a German firm, whose lead drug candidate, CDR132L, works by blocking a particular miRNA in the hope of helping patients with chronic heart failure and cardiac hypertrophy (thickening and stiffening of the walls of the heart).
This year’s prize highlights the continuing trend of recognising discoveries at the molecular level. Recently, several Nobel prizes have been awarded to technologies that have had obvious clinical applications, such as cancer immunotherapy (in 2018) and gene editing (in 2020). It is perhaps no surprise that, as the technology for molecular and genetic research has improved, scientists are gaining ever more insight into cellular function, and are therefore able to make more profound and useful discoveries with them.
Nobel pursuits
For the growing number of researchers around the world who rely on AI in their work, the lasting message of this year’s awards may be a different one: that they, too, could one day nab science’s most prestigious gongs. For his part, said Dr Jumper, “I hope...that we have opened the door to many incredible scientific breakthroughs with computation and AI to come.”" [1]
1. Honouring intelligence. The Economist; London Vol. 453, Iss. 9418, (Oct 12, 2024): 68, 69, 70.
Komentarų nėra:
Rašyti komentarą