Summary: This blog post explores how AI-driven tools are transforming drug discovery by accelerating protein design. We highlight the AI-lab loop approach, which integrates computational and experimental methods for developing new therapeutics, as a way to fully utilize the potential of AI in drug discovery.
Discovery and development of therapeutic molecules has traditionally been a time-consuming and resource-demanding process. It can take over a decade and billions of dollars to bring a single drug to market, with over 90% of candidates failing during the clinical testing stages. This inefficiency stems from the difficulty in identifying molecules that can effectively target diseases without causing significant side effects. For biologics, this challenge is compounded by the issues of stability, deliverability and potential immunogenicity.
Recent advancements in computational de novo protein design (including physics- and AI-based models) have introduced powerful tools for accelerating drug discovery. AI algorithms, such as AlphaFold 3, Evobind, and RFDiffusion, can predict the structure of proteins, deduce molecular interactions, and aid in the design of new drug candidates with desired specifications. These tools have opened up new possibilities for creating highly specific treatments.
The first AI-designed drugs, including both small molecules and biologics, are already in clinical trials. One example is INS018_055, the world’s first AI-designed anti-fibrotic small molecule inhibitor developed by Insilico Medicine, which recently entered into Phase II clinical trial. Another is GB-0669 monoclonal antibody against SARS-CoV-2 developed by Generate Biomedicines, which is currently in Phase 1. The remarkable thing about GB-0669 is that it targets a region previously thought to be undruggable, as well as the fact that the molecule entered the clinic in merely 17 months.
AI tools for the design and discovery of biologics
AlphaFold was the first in silico tool that enabled protein structure prediction with experiment-like accuracy based on sequence. This revolutionary technology, developed by Google’s DeepMind, won the 2024 Nobel Prize in Chemistry. The latest iteration of the algorithm, AlphaFold 3, features expanded capabilities, such as predicting protein-ligand interactions, as well as 50% higher accuracy. Its ability to predict optimal binding sites via a nuanced understanding of protein-ligand interactions holds immense potential for drug discovery.
EvoBind is a recently published AlphaFold algorithm-based tool. EvoBind and AlphaFold serve complementary roles in protein-ligand interaction modeling, but with distinct approaches. While AlphaFold, particularly in its latest iteration, excels in accurately predicting molecular interactions and binding sites using structural data, EvoBind is specifically designed for in silico peptide binder design using only sequence information.
EvoBind applies a directed-evolution approach to optimize peptide binders toward a target protein interface, making it particularly useful when structural binding data is unavailable. Though both tools leverage AI for biomolecular predictions, EvoBind focuses on designing novel peptide binders, offering an open-source solution to aid experimentalists in drug discovery.
RFDiffusion is an alternative approach to protein design, developed by David Baker’s lab at the University of Washington. It takes protein design a step further by generating entirely new protein structures using diffusion models, much like how DALL-E creates images from noise. Unlike AlphaFold, which predicts existing protein structures, or EvoBind, which optimizes peptide binders, RFdiffusion creates novel proteins tailored for specific functions, from drug development to nanomaterials.
By leveraging a guided denoising process, it dramatically reduces the number of experimental tests needed to find viable designs, outperforming traditional methods in tasks like protein binder design, enzyme scaffolding, and symmetric motif generation. Its ability to computationally produce functional proteins with unprecedented efficiency marks a major breakthrough in AI-driven biomolecular engineering.
AI-lab loop approach for de novo protein design
These tools open up new opportunities for drug discovery and design. However, generating safe, functional and effective peptide-based therapeutics still requires extensive lab validation. Even the most powerful AI tools alone are not sufficient to drive the drug discovery process, which is why the combination of computational generation and lab-based validation (referred to as the AI-lab loop) has emerged as a powerful strategy in drug discovery.
The close collaboration between in silico and in vitro-based tools to generate therapeutic agents is called the AI-lab loop. It’s an iterative process where AI algorithms and lab experiments work together in a feedback loop to optimize and refine the design and development of new proteins that are predicted to have the desired features such as structure, stability, activity, and specificity toward a given target.
This approach, when applied to the discovery and development of therapeutic molecules, works with versatile biological modalities including antibodies and their derivatives, TCRs, peptides and proteins of various types, including recombinant, full-size and mini-proteins.
The range of applications of the AI-lab loop approach is equally diverse, including diagnostic tool development (highly specific biomarkers or probes for disease diagnostics), industrial enzymes optimization, design of research tools (innovative proteins for imaging, cell labeling or functional studies), vaccine development, and discovery of minimal active parts of a given molecule.
The proper and effective implementation of the AI-lab loop into research workflows requires specialized expertise in coordinating the computational and experimental stages of the development cycle. Ardigen has partnered with Selvita, which provides wet-lab testing and expertise in a variety of research areas, to offer end-to-end capabilities for AI-based discovery of biological therapeutics.
In an upcoming webinar on March 4, 2025, we will demonstrate examples of successful implementation of the AI-lab loop cycle. If you are interested in learning more about how to effectively implement the AI-lab loop approach to accelerate your drug discovery and development programs, register for the webinar today.
