Naked Scientists: David Baker on Protein Design and the AlphaFold Revolution

The Naked Scientists welcome David Baker, 2024 chemistry Nobel laureate, to discuss how proteins shape life, how predicting their structures evolved from distributed computing to deep learning, and how designing new proteins could tackle pressing problems like plastic pollution and cancer. The conversation navigates from foundational biology to the transformative impact of AI on biotechnology.

• Protein shapes are determined by amino acid sequences and govern life’s machinery
• Early crowdsourced computing projects Rosetta at Home and Foldit engaged the public to predict and design protein structures
• The DeepMind AlphaFold breakthrough leverages large structural databases to predict structures from sequence
• Modern protein design targets real-world problems such as plastic degradation, CO2 fixation, and targeted cancer therapies

Introduction and guest profile

The podcast features David Baker, a Cambridge-based biochemist and 2024 chemistry Nobel laureate, who played a pivotal role in predicting protein shapes and designing novel proteins. The host, Rhys James, frames this episode as part of Naked Scientists season focused on scientific movers who shaped our understanding of biology, chemistry, and biotechnology. The discussion touches Baker’s early life, his late pivot to science, and the central thesis that proteins are the miniature machines of life, with shapes that enable everything from neural signaling to energy capture in plants.

Proteins, structure, and sequence

Baker explains proteins as essential molecular machines whose function hinges on three dimensional structure. Proteins are built from amino acids—20 types—that form linear chains typically 100–500 residues long. The amino acid sequence dictated by genes determines the final folded shape. The surface chemistry of proteins, via amino acid composition, governs interactions with other biomolecules, enabling complex biological tasks. The nanometer scale of proteins means that their shapes are precise and critical to function.

The protein folding problem and early advances

In the 1960s–70s, it was observed that when folded, a protein returns to a defined shape determined by its sequence, implying a sequence-structure code. Yet how the folding code works remained mysterious. Baker’s motivation was to understand self-organization in biology, using proteins as the simplest and most accessible system with thousands of atoms folded into a single functional shape. The challenge lay in modeling thousands of atomic interactions to predict folding, and later, in designing sequences that would fold into a desired structure.

Rosetta at Home and Foldit: crowdsourcing protein science

To tackle the computational complexity, Baker and collaborators launched Rosetta at Home, a distributed computing project that enlisted volunteers to simulate protein folding and design. Participants’ screensavers visualized folding processes, turning abstract calculations into a global citizen science effort. This work gave rise to Foldit, an interactive game that let non-experts manipulate protein structures to find energetically favorable folds. The game included pedagogical levels to teach principles of biochemistry, converting public engagement into real scientific insights. The project also intersected with the SETI community, which shared a scalable computing framework for connecting thousands of personal computers to perform heavy computations.

The deep learning revolution and AlphaFold

The conversation emphasizes the turning point: deep learning demonstrated that a database of known protein structures—created over decades through experimental methods like X-ray crystallography—could be leveraged to learn the rules of protein folding. The AlphaFold program, developed by DeepMind, was trained on thousands of sequences with known structures and could predict the 3D structure of a protein from its amino acid sequence. Baker notes that prior to these advances, scientists spent tens of billions of dollars and decades solving protein structures experimentally; AlphaFold dramatically accelerates this process by providing accurate structure predictions from sequence alone.

From structure prediction to protein design

The interview differentiates structure prediction from design. Baker and collaborators initially used physics-based atomistic models to design new proteins and predict how sequences fold to give a target shape. In the 2000s they demonstrated that a new amino acid sequence could fold into a brand-new shape. The advent of AlphaFold’s predictive power inspired a shift toward design using data-driven methods to generate proteins with specific functions. Baker emphasizes that the suite of methods developed by his group for design, now enhanced by neural networks, enables the creation of proteins with diverse functions, and that these methods have been made openly available for researchers worldwide.

Applications: plastics, CO2, and cancer

The discussion turns to practical applications. Baker’s team is actively designing catalysts that degrade plastics in the ocean, a critical environmental problem. They are also exploring designs to fix CO2, and to target cancer cells with precision, minimizing systemic effects. In addition, the group contributed to the development of vaccines, including a COVID vaccine, demonstrating translational impact from protein design. Baker highlights the potential to combine deep learning with programmable chemistry and the use of non-natural amino acids and cofactors to expand what proteins can do.

Limitations, challenges, and the future

Baker discusses the challenges that AI and design face. Even with deep learning, hard problems in design require iterative cycles: initial designs may fail, revealing where improvements are needed. The integration of traditional physics-based modeling with AI opens possibilities for designing catalysts using unnatural amino acids or cofactors. The host and guest discuss the importance of collaboration, the tight-knit lab environment in Seattle, and Baker’s weekend escapes to the mountains for balance and inspiration.

Closing reflections

The episode closes with reflections on the pace of discovery, the intersection of science and medicine, and the enormous potential of protein design to address sustainability and health. The hosts acknowledge the Nobel Prize recognition as a milestone that accompanies continued opportunities to advance science and improve human health through design and AI-enabled discovery.