Protein design is a daunting task, akin to searching for a needle in a haystack. The haystack represents the vast search space of possible protein designs, with each piece of straw corresponding to a unique combination of amino acids and structural arrangements. The needle, on the other hand, represents the specific design that works – the one that binds with high affinity, folds correctly, and exhibits the desired biological function. With 20 standard amino acids and countless possible combinations, the haystack is enormous, and the needle is tiny. But what if we could use a powerful tool to guide our search, to focus on the most promising areas of the haystack and narrow down the possibilities? That’s where RFdiffusion comes in, a cutting-edge generative AI algorithm that’s revolutionizing the field of protein design.
Protein-protein interactions are crucial to virtually all biological processes, from signaling pathways that control cell growth, differentiation and apoptosis, to metabolism and modulating immune responses, and many other vital functions. The multifaceted roles of protein-protein interactions make them a significant reservoir of potential drug targets, offering new mechanisms of action that are attracting considerable attention. The ability to understand these interactions and design protein binders is not limited to therapeutic drug discovery. In research and industry applications, binder proteins can be useful tools for in-vitro and imaging diagnostics, and biosensing and biocatalysis.
The goal of designing a protein binder is to discover or refine protein sequences that exhibit specific binding capabilities, often to improve upon existing interactions or create entirely new ones. The problem is that the design space is mind bogglingly massive, with most designs lacking the desired function in the biological context of interest (organism, temperature, pH) or not folding into the desired conformation. Before the arrival of recent AI models and developments, the design of protein binders was a significantly more challenging, time-consuming, and often less predictable process. Scientists relied heavily on a combination of experimental techniques, computational methods such as structure-based design with homology modeling and docking, or antibody modeling tools focused on framework region and specialized loop modeling for CDRs, and evolutionary approaches.
Over the last decade this has been changing with the work of Professor David Baker, and particularly with the publication of RFdiffusion, a ‘general deep-learning framework for protein design. in 2023. Its momentous significance was cemented by the award of the 2024 Nobel Prize in Chemistry to Prof Baker for his work in computational protein design. This award was shared jointly with Demis Hassabis and John Jumper from Google DeepMind for protein structure prediction.
Generating Proteins Binders with AI: RFdiffusion
At its core, RFdiffusion leverages a sophisticated generative AI approach. Imagine starting not with a defined protein, but with a diffuse cloud of atomic coordinates—essentially structural ‘noise.’ The algorithm then meticulously ‘denoises’ this initial state through a series of iterative steps. In each step, it refines the positions of these atoms, guiding them towards forming a coherent, stable, and biologically plausible protein structure. This process is akin to a sculptor gradually revealing a detailed figure from a rough block of stone, or a photographer sharpening a blurry photograph into a crisp, clear image, with each refinement bringing the final, intended design into focus.
This powerful generative capability opens doors to tackling a wide array of biotherapeutic design challenges. For example, RFdiffusion can be employed to design novel biologics engineered to bind and neutralize viral targets. In the realm of protein-protein interactions, whether involving antibodies or other systems, it can generate new structural scaffolds aimed at optimizing binding affinities or bolstering the stability of complexes. Furthermore, the algorithm shows promise in designing enzyme-based therapeutics, such as those capable of degrading specific metabolites to address metabolic diseases. The applications extend beyond medicine. RFdiffusion holds vast potential for custom-designing proteins for industrial and biotech purposes, from creating enzymes that drive specific chemical reactions with high precision to developing proteins that remain functional under extreme environmental conditions like varied temperatures or pH levels.
RFdiffusion became available to users of BIOVIA Discovery Studio Simulation in the 3DEXPERIENCE® Cloud through the Generate Protein Scaffold protocol in 2024. This implementation provided users with a powerful tool for AI-driven protein design, enabling them to access the algorithm and tackle a wide range of protein design problems, including enzyme active site scaffolding, symmetric motif scaffolding, oligomer design, and protein binder design. The latest version of the Generate Protein Scaffold protocol significantly enhances protein binder design by enabling users to easily define specific ‘hotspots’ on the target protein that new scaffolds must engage. By specifying these critical residues – areas known or predicted to be vital for interaction – users can direct RFdiffusion to focus its generative power on designing binders that make meaningful contact precisely where it matters. This targeted approach dramatically increases the likelihood of generating binders with high affinity and specificity, while also conserving computational resources by concentrating the search within the most promising regions of the protein-protein interface.
See how BIOVIA Discovery Studio Simulation leverages RFdiffusion and generative AI to accelerate protein binder design.
Multi-step AI Workflow for Proteins Binders
The journey from concept to a promising candidate involves several key AI-driven steps. First, Generate Protein Scaffolds leverages RFdiffusion to create an innovative 3D scaffold. These structural blueprints are then passed to the Generate Protein Sequences protocol. Powered by sophisticated models like ProteinMPNN, this protocol takes the designed shape and predicts optimal amino acid sequences – the ‘beads’ for the string – intended to fold correctly and exhibit the desired binding function. Crucially, to validate these novel sequences, they are subsequently fed into the Predict Protein Structures protocol. This step utilizes leading-edge methods like OpenFold and AlphaFold2 to computationally predict the 3D structure that the newly designed amino acid sequence will adopt. This provides an essential in-silico validation confirming that the designed sequence is indeed likely to adopt the intended fold and bind the targeted interface. This comprehensive, multi-step AI workflow – from initial scaffold generation by RFdiffusion, to sequence design with ProteinMPNN, and finally to structure verification using models like AlphaFold2 – is what transforms the complex challenge of protein binder design into a significantly more manageable and effective task for researchers, ultimately accelerating the biologic development towards experimental validation.
Empowering Discovery: The Future of Protein Binder Design with Generative AI
The biotherapeutics design tools in BIOVIA Discovery Studio Simulation include an ever-growing array of powerful AI tools for molecular modelers and biologists to help in their efforts to design effective protein binders. Particular attention is also being paid to the user experience of these tools to make them more accessible and straightforward for our users. That will continue in future developments, to create workflows that in combination with existing physics-based methods will allow users to rapidly explore many more possibilities in silico before arriving at the best candidates that are ready to become successful commercial biotherapeutics or biologics to be used in diagnostics, and biosensing and biocatalysis.
📩Want to find out the latest news about BIOVIA events, customer stories, blogs and more? Join the monthly newsletter today!
Tien Luu is the Product Manager and Principal Scientific Specialist at BIOVIA Dassault Systèmes, managing aspects of life science modeling and simulation predictive science solutions. She began her career as a Scientific Specialist after receiving her Ph.D. in Chemistry from the University of Wollongong, Australia. Since joining BIOVIA 22 years ago, Tien’s passion and excitement for helping customers accelerate their R&D have been steadfast and she is committed to improve customer experiences and product successes. In addition, she continues to be an active leader in scientific software product development, new program initiatives, and new applied and fundamental scientific research.