How AI Helped Solve One Of Biology’s Biggest Problems, And Why It May Just Be the Start

AlphaFold’s breakthrough: A new era for protein science begins

In December 2020, John Jumper from Google DeepMind presented AlphaFold2, an AI tool that could predict protein structures with over 90% accuracy – solving a 50-year-old challenge known as the protein folding problem and marking the biggest AI breakthrough in science to date.

The protein folding problem takes center stage: For decades, scientists sought to crack the code of how a protein’s amino acid sequence determines its folded 3D shape, which in turn defines its biological function:

• In the 1950s, Christian Anfinsen hypothesized that a protein’s amino acid sequence contains the information needed to predict its structure, sparking efforts to solve the “protein folding problem.”
• The Critical Assessment of Structure Prediction (CASP) experiment, founded in the 1990s, became a proving ground for computational approaches, but progress remained incremental.
• By 2020, the Protein Data Bank contained over 140,000 experimentally determined protein structures, providing a wealth of training data for AI algorithms.

AI revolutionizes protein structure prediction: AlphaFold2’s success relied on key AI breakthroughs and unprecedented computing power:

• Advances in deep learning, inspired by the neural networks of the human brain, enabled the development of highly sophisticated algorithms.
• DeepMind harnessed Google’s massive computational resources to train AlphaFold2 on the Protein Data Bank’s extensive dataset.
• Under the leadership of John Jumper, the algorithm incorporated cutting-edge transformer architectures to predict protein structures with unrivaled accuracy.

Shock, excitement, and existential questions: AlphaFold2’s breakthrough sent shockwaves through the scientific community, eliciting a range of reactions:

• Some feared that the AI would make structural biologists obsolete, while others were ecstatic about the potential to accelerate their research.
• The media sensationalized AlphaFold2’s potential to revolutionize drug discovery, but many experts cautioned that significant challenges remained.
• The success of a newcomer over established experts sparked debates about the nature of scientific discovery and the role of AI in advancing knowledge.

Pushing beyond structure prediction: While AlphaFold2 excels at predicting the structures of individual proteins, it has limitations when it comes to modeling the dynamic complexity of living cells:

• Proteins constantly interact with other molecules, altering their shapes and functions in ways that AlphaFold2 struggles to capture.
• Shape-shifting and intrinsically disordered proteins, which play critical roles in biology, pose significant challenges for the AI.
• DeepMind’s AlphaFold3 and David Baker’s RoseTTAFold aim to predict the structures of protein complexes, but their accuracy remains limited compared to AlphaFold2’s single-protein predictions.

Toward a new paradigm for biological discovery: AlphaFold2’s success has kindled deep questions about the future of AI in science and the very nature of scientific inquiry:

• Some worry that relying on AI predictions without understanding the underlying processes could undermine the pursuit of fundamental knowledge.
• Others argue that if AI can solve complex problems and cure diseases, the means by which it arrives at those solutions may be less important.
• As the field grapples with these philosophical questions, researchers are exploring ways to harness AI’s power while preserving the essence of scientific discovery.

Analyzing deeper: AlphaFold2’s groundbreaking success in solving the protein structure prediction problem marks the beginning of a new era in which artificial intelligence is poised to transform key areas of biological research. However, significant challenges remain in modeling the dynamic complexity of living systems, and the rise of AI has sparked deep questions about the future of science itself. As researchers seek to harness the power of AI while grappling with its limitations and implications, they are forging a new paradigm for discovery – one that will likely continue to evolve in surprising and profound ways, just as the protein