DenovAI is the second start-up to launch from the global pharma consortium AION Labs – an organisation that aims to improve the drug discovery and development process through the use of AI and other computational technologies. DDW Editor Reece Armstrong, speaks to Kashif Sadiq, Founder of DenovAI Biotech about the importance of computational methods in this field.
RA: Could you tell me about the launch of AION Labs’ second start-up DenovAI?
KS: I have been involved in the field of computational molecular biophysics for many years and more recently in protein design with my work at the European Molecular Biology Laboratory (EMBL). My team and I there focused on developing a new algorithm which is capable of performing de novo protein design to create entirely new proteins that don’t exist in nature but yet still fold into stable three-dimensional proteins.
This brought us to intersect with AION Labs, which had issued a call for applicants to launch a new company leveraging AI-based solutions for the computational de novo design of high-affinity antibodies for targeted therapies. Recognising how the technology we had developed with EMBL complemented AION Labs’ objectives, we applied for the challenge, resulting in the formation of DenovAI Biotech as a company focused on developing an AI-powered biophysics solution for de novo antibody discovery.
Through a process of technology transfer, the technology we developed at EMBL is now licensed to DenovAI through EMBL Enterprise Management Technology Transfer GmBH (EMBLEM), a wholly owned commercial subsidiary of EMBL. Excitingly, we are now in position to advance in collaboration with all our AION Labs partners.
RA: How laborious and expensive are current antibody discovery efforts?
KS: In short, incredibly so. Most antibody discovery outfits rely on large experimental libraries comprising up to a billion antibody sequences. These repositories need to be sifted through to find out if any of the sequences hit on a specific target. The process is not just incredibly work-intensive, costly and time-consuming, but often results in failure.
RA: What kind of platform is DenovAI building to help with antibody discovery?
KS: The premise of what we are doing at DenovAI is to invent methods that bypass the current laborious process of antibody discovery, speeding up the process with a neat, AI-driven in-silico solution. To that end, we are leveraging the technology that we developed at EMBL in the de novo protein design space, building on it to solve the greater challenge of designing large flexible proteins, specifically antibodies, to be used as therapies.
The platform we are building will be able to rapidly search a vastly larger sequence space of antibodies, dwarfing existing experimental libraries, and return predictions for several designed antibody proteins – not only their sequence, but also their structure bound to the intended epitope. This will significantly shorten the timescale to discovery to a matter of days or weeks, as opposed to the current standard of months to a year.
RA: Why is this such a novel endeavour?
KS: What makes this truly novel is that we are attempting to apply the solution to any given epitope. With current AI approaches, there is a lot of work taking place in the generative AI space. When applied to antibodies, this amounts to developing a vast number of different sequences, some of which could also be viable antibodies.
At DenovAI, we are also considering the epitopes and working to develop a solution that could be applied to any epitope of interest. This would be truly foundational, as it could galvanise the entire industry, enabling us to design antibodies from scratch that could be applied to almost any target involved in a disease.
This would allow us to target not just those diseases whose protein targets are already known, but also possess a repository of antibodies for diseases and their corresponding targets that have not yet been identified. That would truly stimulate the industry and encourage further drug discovery efforts applying our technology to address the thousands of diseases that are not currently being investigated.
RA: How is DenovAI building on recent advances in protein structure prediction and AI to develop its platform?
KS: The advent of AlphaFold has been a gamechanger in structure prediction, providing a tremendous opportunity to solving the inverse problem of protein design. For decades, one of the great challenges in biology has been how to predict the folded structure of a protein sequence. AlphaFold was one of the first software solutions capable of solving this problem, enabling us to predict the three-dimensional structure of a protein with unprecedented accuracy.
When AlphaFold launched, we were working on a solution to the inverse problem of how to predict the correct sequences that will result in the desired design features of a new protein. We were able to leverage this to develop a wider algorithm for de novo protein design and create completely new proteins that do not exist in nature. This is the stage at which we intersected with AION Labs and its particular focus on antibody development.
RA: Additionally, how has the increased availability of antigen-antibody structures helped with this project?
KS: AlphaFold was designed to be a general structure prediction tool for any protein and has at its disposal hundreds of thousands of experimental structures. The key challenge with antibodies is that there are nowhere near the same number of available structures. Having access to more data is thus essential to the success of any antibody development project. Thanks to the AION Labs alliance, we will be able to access data that is not available in the public domain. This will enable us to galvanise our algorithms to become even more effective, with the additional datasets providing a solid foundation on which to train our models.
RA: Could you describe the importance of computational molecular biophysics in antibody discovery and how this is being incorporated into the company’s platform?
KS: This is another area where we believe we can provide real differentiation. Understanding protein function from structure is akin to trying to predict the range of capabilities of motion for a human based on a statue of that person. Dynamics are thus crucial to understanding protein function and computational molecular biophysics is the field that applies to understanding the dynamics of biomolecules, including proteins.
This is where we have made considerable advances, by working to understand not just what protein structures look like, but how they change conformations over nanoseconds to milliseconds of their life. Those conformational changes affect and are intimately linked to their function. Antibodies, in particular, have some very flexible regions. We plan to go one step further and design proteins not on the basis of structure alone, but on the underlying function that may emerge from the dynamics. Thus, we believe computational molecular biophysics will be integral to accurate antibody discovery.
RA: Could the platform DenovAI is developing be applied to other drug discovery areas outside of antibodies?
KS: As a matter of fact, we initially developed the platform to address small proteins before moving on to large antibodies. Small proteins are also particularly important therapeutic modalities and there is still considerable appeal for them in certain areas where they can prove more effective than antibodies. On the whole, antibodies are performing well, but there are niche areas where smaller proteins could outperform them. Thus there is definitely a case of building capabilities in that direction, which is something we are also looking to work towards.
DDW Volume 24 – Issue 4, Fall 2023
Kashif Sadiq is Founder and CEO of DenovAI Biotech, an AION Labs company developing an AI-powered biophysics solution for de novo antibody and small protein binder discovery. An expert in computational biophysics and protein design, Sadiq holds a BA MSci in Natural Sciences from the University of Cambridge and a PhD in Theoretical and Computational Biophysics from University College London.