In November of 2024, a new iteration of a revolutionary technology was released, one that took the biotechnology and pharmaceutical industries by storm. With its technological brilliance and a growing myriad of impact stories to demonstrate its real-world significance, the development was none other than Google’s AlphaFold – specifically, AlphaFold 3.
Developed by artificial intelligence research companies DeepMind and Isomorphic Labs, AlphaFold is an AI model that can predict the structures and interactions of molecules: things like proteins, RNA, DNA, etc.
Since the initial release of the first model in 2018, the technology has undergone continuous development, ultimately leading to what we see today. From winning the 2024 Nobel Prize in the chemistry category to acting as a critical tool for individual researchers and large pharma companies alike, recognition of the technology’s importance only grows.
To understand what makes the technology so profound, we first need to understand something called the “protein folding problem” – one of the largest problems in biological research. Protein molecules consist of incredibly long and complicated chains. Consisting of many organic compounds, specifically amino acids, that can all interact with one another in endless different ways, there’s an astronomical amount of possible configurations a protein could adopt. And since the molecular structure of a protein plays the biggest role in our understanding of its function, research within the field used to be incredibly difficult. Figuring out a single protein’s shape could take years of experimentation, and the experimental methods that did exist were very time and resource intensive. This is where AlphaFold came in.
By the mid-2010s, DeepMind had already left its mark on the AI industry. The company was gaining recognition due to its mastery over complex games like Go and chess. In fact, at the time, their AI model had gotten good enough to beat the world Go champion, Lee Sedol. After this achievement, the team decided to set their sights on a much larger project. Google DeepMind CEO Demis Hassabis said, “I was feeling like it was time to tackle something really hard in science because we had just solved more or less the pinnacle of games AI. I wanted to finally apply the AI to real-world domains. That’s always been the mission of DeepMind: to develop general-purpose algorithms that could be applied across many, many problems. We started off with games because it was more efficient to develop AI and test things out in games for various reasons. But ultimately that was never the end goal.”
And so, the company moved onto AlphaFold. In 2018, the earliest model of AlphaFold was completed. Soon enough, through its revolutionary performance in the Critical Assessment of protein Structure Prediction competition, AlphaFold signalled an entirely new era in protein structure prediction. However, its next model made even more headlines.
Released in 2020, AlphaFold 2 came equipped with new architecture, training methods, and a totally unprecedented accuracy. In fact, after it won the biennial Critical Assessment of Structure Prediction with exceptionally high scores, CASP organisers proclaimed that the protein structure prediction problem “[had] been largely solved for single proteins”.
But how did it even get here? Well, using a protein data bank consisting of 215,000 entries, the AI was taught to learn pattern identification within a large dataset. And after being trained on existing data, the technology could now predict properties of new proteins, and even predict previously unknown ones. Additionally, model 2 was even equipped with confidence metrics. If aware that its prediction might not be a hundred percent accurate, the AI would give a confidence score that reflected the potential accuracy.
Now, finally, we have AlphaFold 3 – the most recent version.
This model was initially announced on May 7, 2024. At a briefing, Hassabis announced the breakthrough. “Biology is a dynamic system and you have to understand how properties of biology emerge through the interactions between different molecules in the cell. You can think of AlphaFold 3 as our first big step towards that.”
Model 3 essentially solved the problem of predicting the structures of almost all biological molecules and modeling the interactions between those molecules. Its performance is revolutionary, showing us over a 50% improvement compared to previously existing methods of prediction.
“AlphaFold continues to get better, and increasingly more relevant for biological investigations,” said Paul Nurse, the chief executive and director of the London-based Francis Crick Institute. “This third version will enable increased accuracy in predicting the structures of complexes between different macromolecules, as well as associations between macromolecules, small molecules and ions.”
Now, while only Isometric Labs can use AlphaFold for commercial purposes, scientists can access the majority of its capabilities for free through something called AlphaFold Server. Given an input list, the research tool generates their joint 3D structure, revealing how they all fit together. The ability to predict protein binding is critical to understand human immune response and design new antibodies, making it indispensable for biomedical research.
Right now, AlphaFold 3’s uses are diverse and varied. According to Google DeepMind, up to a billion research years have been saved by the technology, and low to middle income countries have over 600,00 AlphaFold users. From being used to find potential Parkinson’s treatments, to increasing honeybee’s chances of survival, “the impact of AlphaFold can be seen in nearly every field of biology”.
Thus, AlphaFold 3 represents a defining moment in the intersection of AI and, not just biology, but science overall. Individual researchers, labs, and students around the world now have access to cutting-edge tools that can accelerate crucial discoveries and treatments. As AlphaFold 3 continues to garner more and more use, we see in real-time the immense potential of AI in unlocking entirely new possibilities in science and research applications.
Written by Saanvika Gandhari
