By Sean McClain, Absci Founder and CEO

Synthetic biology offers a diverse collection of tools to help accelerate existing processes in pharma. But these capabilities remain scattered and difficult to integrate. Today, a new generation of synbio-native companies offers drug makers a more cohesive approach by flattening the tech stack in drug discovery and manufacturing.

But synbio alone cannot transform the existing paradigm from drug discovery to drug creation. One more critical ingredient is needed to reduce preclinical development times and increase the clinical probability of success: artificial intelligence.

Following success in areas such as computer vision and natural language processing, artificial intelligence (AI) is now advancing drug discovery. First applied to small molecules, AI drug discovery has now come to biologics like antibodies, protein therapeutics, and gene therapies. These therapeutic modalities have high levels of target specificity as well as broad applicability across medical research. The marriage of AI and synthetic biology unlocks the potential of both of these exponential technologies to create better biologics for patients faster.

A critical component of the union of synbio and AI is data at scale. Vast quantities of high-quality data are critical for training accurate and effective AI algorithms. Until recently, the lack of biological experimental data has prevented AI from being used effectively in developing biologic drug candidates. But synbio enables us to overcome the biological data challenge with high-throughput wet lab technology. Today, advanced synbio screening means drug developers at Absci can create and experimentally validate millions of AI-generated designs per week.

De novo antibodies

Researchers at Absci have recently shown how we can leverage vast biological data with generative AI capabilities to design and validate antibodies with properties that could never be found with traditional discovery approaches. Using a zero-shot drug discovery method, it’s possible to design antibodies to bind to specific targets without using any training data of antibodies already known to bind to those specific sites. The result: de novo antibody designs unlike any found in existing antibody databases.

It’s important to point out that this was not just another in silico library — these AI-designed antibodies are produced and validated in the wet lab. The entire design-build-test-learn cycle takes about six weeks. This is a testament to the combined power of synbio and AI.

De novo antibodies show the ultimate potential of synbio and AI in biologic drug discovery. But antibodies are just one example of how AI is impacting the pharma tech stack at large.

Target identification

We can also apply synbio and deep learning techniques to identify new drug target /antibody combinations. Reverse immunology is grounded in natural immunology, beginning with real-world samples from patients who display exceptional immune responses. Using the RNA sequences of their immune proteins as antibody blueprints, we can computationally re-assemble fully human antibody sequences in silico. We can then produce these antibodies in the lab and use high-throughput proteomic screening methods to determine their target antigens in an experimental process called “deorphaning.” At the end of this reverse immunology process, we have the immunologic target paired with a fully human monoclonal antibody that can be used as a starting point for drug development.

Lead optimization: Safety, efficacy, and developability

Lead optimization for biologics can be viewed as fine-tuning the best antibody candidates in three areas: efficacy, developability, and safety. In terms of safety, AI models can be used to optimize antibody candidates on what we call a Naturalness score. This metric scores antibody variants for similarity to hundreds of millions of natural immunoglobulin sequences in our training set. We’ve demonstrated that the Naturalness score is inversely associated with immunogenicity, which means antibodies with high Naturalness scores may have a lower likelihood of triggering unwanted immune responses. By optimizing Naturalness scores in the drug creation process, we expect to mitigate downstream developability and immunogenicity issues by prioritizing highly natural antibody candidates.

How antibodies attach to target proteins in the body can be critical to efficacy (e.g., how they attach to each other and how sequence changes to the antibody might strengthen their binding). A number of AI tools are available to predict how two proteins will interact, using ML methods like Monte Carlo or fast-Fourier transform to score the energetic favorability of two docked structures. These same methods also can be used to screen candidate antibodies for properties that are important to developability, including stability and solubility. When incorporated into an integrated drug creation platform, these AI tools help filter antibody candidates and narrow the search space.

Codon optimization

Synbio and AI have been combined to solve another longstanding challenge in the field: codon optimization. Simply put, codon optimization is the process of finding the perfect DNA sequence to maximize the production of the desired protein therapeutic in a host cell. There are a lot of tools out there, but we’ve now developed an AI-based codon optimization solution that outperforms commercially-available algorithms. The results suggest the AI model had learned fundamental rules governing codon optimization. AI-based codon optimization could theoretically be applied across protein classes to save significant time and money by maximizing protein production at scale.

Protein structure and function

The deep learning-based approaches for structure prediction, exemplified by AlphaFold, are having a transformative impact in drug discovery. Going from structure to function is a holy grail in pharma, and a number of new AI approaches are exploring the exciting frontier. Already, scientists have immediate clues to better understand disease mechanisms and design more effective drugs that target specific proteins.

This is especially true for biologics, where AI-based structures are unlocking the target specificity and broad applicability of antibody approaches to disease treatment. The protein structures of antigens are an important input to de novo antibody designs, especially where the crystal structures of antigens are unknown. This area is evolving rapidly.

From drug discovery to drug creation: A more cohesive approach

In the coming years, AI promises to have a very broad impact in pharma. In clinical trials, it can assist in identifying suitable patients, monitoring their progress, and analyzing data. In drug manufacturing, it can improve the production and quality control of drugs. And in personalized medicine, I’m hopeful that AI will one day soon predict patient responses to specific treatments and tailor therapies accordingly.

Today, Absci is focused on how AI can impact drug discovery. Synbio and AI have combined to create a more cohesive approach to drug discovery and manufacturing. Synbio's high-throughput wet lab technology enables the generation of vast biological data, which trains the AI models to design and validate antibodies with properties that traditional discovery approaches could never find. These technologies also hold promise in target identification, lead optimization, codon optimization, and the structure-function challenge, among other applications in biologic drug discovery. By combining AI and synbio in an integrated platform, we are moving the pharma model from drug discovery to drug creation, and toward our vision to deliver breakthrough therapeutics at the click of a button for everyone.