Rise of AI in Biomedicine: AlphaFold

Awarding the Nobel prize in Chemistry to the inventors of AlphaFold marks an important milestone for the impact of AI in Life Sciences.

The title may seem slightly misleading, as artificial intelligence (AI) has been used in biomedicine for decades. Historically, however, its applications have generally led to incremental advances rather than transformative breakthroughs. The introduction of AlphaFold, a revolutionary tool for protein structure prediction, has marked a turning point for AI in the biomedical field.

Predicting protein structures from the nucleotide sequences of genes has been a longstanding challenge in biology. Solving this problem has profound implications for drug design. Understanding a protein's structure is one of the foundational steps in designing chemical drugs that specifically target it. Additionally, predicting the structure of therapeutic proteins under development—such as monoclonal antibodies—is crucial for their success as drugs.

For many years, experimental techniques like X-ray crystallography have been the gold standard for determining protein structures. However, these methods are not always practical, as they demand substantial time and resources. Furthermore, certain types of proteins, such as membrane proteins or multi-protein complexes, are particularly difficult to analyze using these experimental approaches. In diseases like cancer, where mutations can alter protein structures, it is essential to capture these changes for effective drug design.

Computational biologists have attempted to address this problem for decades, but the immense complexity of protein structure prediction has rendered most solutions ineffective—until the arrival of AlphaFold. Developed by Google DeepMind and first introduced in 2018, AlphaFold revolutionized the field. The initial version utilized a Convolutional Neural Network (CNN) architecture. AlphaFold 2, launched later, employed a transformer-based architecture similar to that used by ChatGPT. The latest iteration, AlphaFold 3, combines transformer and diffusion-based models, extending its capabilities to predict not only protein structures but also those of DNA, RNA, post-translational modifications, and ions.

The scientific impact of AlphaFold is monumental. To illustrate, consider the 2024 Nobel Prize in Medicine, awarded for the discovery of micro-RNA—first documented in 1994. It took 30 years for the Nobel Committee to recognize the significance of this discovery. In stark contrast, AlphaFold’s inventors, Demis Hassabis and John Jumper, received a Nobel Prize in Chemistry in 2024 for protein structure prediction just six years after its introduction, highlighting the transformative nature of this achievement.

AlphaFold exemplifies the extraordinary potential of AI in the biomedical sciences. It also demonstrates that AI extends far beyond the Large Language Models (LLMs) like ChatGPT that have dominated public discourse. While LLMs exhibit remarkable cognitive abilities in text processing, much of scientific research relies on quantitative data. Tools like AlphaFold have the potential to drive dramatic advancements in translational medicine, transitioning from incremental progress to revolutionary breakthroughs. As AI continues to evolve, we can expect its impact on biomedicine to accelerate, shaping the future of healthcare and drug development.

Note:

Image generated with ChatGPT.

Additional Resources:

AlphaFold database homepage: https://alphafold.ebi.ac.uk/ 

AlphaFold publication in Nature: https://www.nature.com/articles/s41586-021-03819-2 

An introductory article for AlphaFold: https://medium.com/the-21st-century/alphafold-ai-for-protein-structures-prediction-3395bc44a00e 

Another article for learning more about AlphaFold: https://towardsdatascience.com/deepmind-alphafold-the-protein-folding-problem-a331d9ced746/