The latest on Isomorphic Labs’ plans to use AI models to revamp drug discovery

“Nearly all diseases stem from a molecular mechanism going wrong,” writes Sergei Yakneen, chief technology officer of Alphabet’s Isomorphic Labs in a recent blog post. “Whether it’s an infectious agent disrupting our cellular machinery or a breakdown in processes like DNA repair, identifying the right protein and finding molecules to interact with it is crucial for drug discovery.” Yet Yakneen points out the sheer scale of this challenge: “The number of potential molecules to consider is astronomical, making the search for effective treatments complex, time-consuming, and expensive.”

The role of AlphaFold 3

The latest salvo in Isomorphic Labs’ arsenal is AlphaFold 3, the AI model it co-developed with Google DeepMind that can accurately predict the 3D structure of proteins. That capability has the potential to allow researchers to identify promising drug candidates more quickly and with greater precision.

An article published in July Nature Structural & Molecular Biology highlighted AlphaFold 3’s ability to advance researchers’ understanding of protein structures and their interactions. Titled “AlphaFold3 takes a step toward decoding molecular behavior and biological computation,” the article describes this iteration of AlphaFold as a “breakthrough in predicting the 3D structures of complexes directly from their sequences”

In a blog post also Shweta Maniar, Global Director of Healthcare and Life Sciences at Google Cloud, Yakneen emphasizes that while AlphaFold 3 is a powerful tool, Isomorphic Labs’ vision extends beyond just protein structure prediction. Isomophic Labs and Google DeepMind are developing a comprehensive AI platform that addresses multiple facets of drug discovery, including:

• Target identification: AI models can be used to analyze large amounts of biological data to identify the specific proteins involved in a disease process. That translates into more precise targets for drug development.
• Drug-target interaction prediction: By understanding how potential drugs bind to and interact with target proteins at a molecular level, AI can help predict the efficacy and specificity of drug candidates. That can curb the reliance on costly and time-consuming experimental screening.
• Drug efficacy and safety prediction: In addition, AI models learn to predict the potential effectiveness and side effects of drug candidates based on their chemical structure and predicted interactions within the body. That can potentially streamline the drug development pipeline, which is one of the core promises of AI in the space.

Growing research traction

AlphaFold 3, which launched in May 2024, is the subject of a considerable amount of research. Examples include a review titled “Review of AlphaFold 3: Transformative Advances in Drug Design and Therapeutics,” which highlights the model’s accuracy and power compared to its predecessor, AlphaFold 2. The review also notes AlphaFold 3’s ability to predict protein structures in seconds, a task that would take humans years to complete.

Another recent study, “Benchmarking AlphaFold3’s protein-protein complex accuracy and machine learning prediction reliability for binding free energy changes upon mutation” evaluates AlphaFold 3’s predictions using the SKEMPI 2.0 database, which contains data on 317 protein-protein complexes and 8,338 mutations.

A third 2024 study, titled “Evaluation of AlphaFold 3’s Protein–Protein Complexes for Predicting Binding Free Energy Changes upon Mutation,” also explores AlphaFold 3’s ability to probe complex structures using the SKEMPI 2.0 database.

One of the most influential papers on AlphaFold 3 was the first. Published in Nature in May 204 and titled “Accurate structure prediction of biomolecular interactions with AlphaFold 3,” the paper described AlphaFold 3’s ability to predict the structure of a range of biomolecular complexes, including proteins, nucleic acids, small molecules, and various modifications. The paper noted how AlphaFold 3 can achieve substantially improved accuracy over many previous specialized tools across different types of molecular interactions.

This foundational work paved the way for a new era in computational drug discovery, where AI models like AlphaFold 3 can rapidly and accurately predict complex biomolecular interactions. 

In his post, Yakneen emphasizes that Isomorphic Labs’ AI-driven platform is not designed to replace experimental validation but rather to guide and accelerate the process. “By reducing the experimental burden,” he writes, “we can pave the way for faster, more efficient development of new treatments and cures.”

PDB 7RCE, an engineered homing endonuclease bound to DNA. Image made in AlphaFold 3. Abramson, J et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature (2024)