Antibody Therapy in the Era of AI: A TechBio Approach
Discover how Artificial Intelligence (AI) is transforming drug discovery, from small molecules to biologics. Explore the use of AI in drug development and accelerating discovery, with a focus on novel antibodies.
Listen to this article
Audio copyright: immunitoAI
The Future of Therapeutics
Antibodies are an integral part of the body's natural defence machinery. When the immune system detects foreign substances such as bacteria, viruses, and pathogens, it produces proteins called antibodies. These antibodies defend the body by specifically recognising and binding to target molecules, called antigens. This antigen-antibody interaction triggers various immune responses in the body to destroy the invaders.
Antibodies are Y-shaped molecules, and the tips of the Y form the antigen-binding sites. Antigens have a specific region on their surface that is recognised by the antibody, making this interaction very precise. The strength of this antigen-antibody binding is known as affinity. These two properties- specificity and affinity, make antibodies an essential part of the biologics landscape for targeted therapy.
Conventional chemotherapy, aimed at eradicating rapidly-growing cells, lacks specificity. These drugs cannot distinguish between cancer cells and healthy fast-growing cells, leading to debilitating side effects. In contrast, the specificity of antibody therapy allows it to activate the immune system to selectively target and destroy cancer cells, sparing healthy tissues.
Y-shaped antibodies binding to cancer cell (in purple) and recruiting immune cells (in yellow).
Scientists can produce antibodies against a particular target molecule associated with a disease, in the laboratory. These antibodies can then be administered to patients as drugs for targeted therapy. Currently, 5 of the top 10 selling drugs are antibodies. According to a report by EvaluatePharma from April 2022, antibodies are expected to become the largest sector for cancer drugs by 2030.
The technology to generate antibodies in the laboratory was established in 1976, and the first antibody drug was approved for clinical use in the late 80s. Advancements in technology and our understanding of immunology have evolved the method over years. Yet, antibody discovery remains a laborious and resource-intensive process, and this raises the cost of antibody therapy. Compared to small molecules, antibodies are a newer and underexplored form of therapeutics. Hence, researchers and pharmaceutical companies are focused on finding new approaches of antibody discovery, to tap the true potential of antibody therapy.
The classical method involves using animals as a source of antibodies. Usually a mouse (or other non-human mammal) is injected with the chosen target antigen to stimulate an immune response. This process is called immunisation. The antibody making cells are then collected and can be grown in the lab to produce the desired antibody in large quantities. The antibody specifically binding to the target antigen is extracted and purified. At this point, we have an antibody as a lead molecule for further development.
The antibodies undergo extensive characterisation to ensure they have desired properties like stability, solubility, non-toxicity, non-aggregation etc. These properties are referred to as ‘developability’ of an antibody. The molecule is modified in a hit-and-trial fashion to find the sweet spot of optimising these attributes while preserving its therapeutic properties. This is an iterative process. It is only after this stage that the antibody proceeds for next phases of drug discovery: preclinical testing, clinical trials and regulatory approval.
Read more about the phases of drug discovery and the applications of Artificial Intelligence in the field.
The sequence of antibody development in the lab
Making Discovery Faster, Cheaper and Efficient
The conventional process of antibody drug discovery meets many challenges along its path. Usually, a large population of animals needs to be immunised, out of which not all survive. Sometimes, the surviving population has to be injected again to produce the desired antibody. Since these antibodies are derived from animals like mice, they can potentially be recognised by the patient’s immune system as foreign. This property of therapeutic antibodies to trigger an immune response in the patient's body against the drug itself, is called immunogenicity. This reduces the effectiveness of the treatment and can lead to adverse reactions.
Dr. Abhishek Mathur, Senior VP and Head of R&D at Enzene Biosciences points out another limitation. “In the traditional antibody discovery process, post-immunisation of the animal, the antibodies are generated by the pertinent immune system. While we can get large antibody panels through this approach that may eventually result in a potential drug candidate, it is a lengthy and cumbersome process and we are still limited by the sequences that the immune system could generate. Antibody discovery is a statistical game and if we could generate the de novo sequences synthetically through well defined educated algorithms in place, we could significantly increase the chances of success in the clinic and importantly would not be limited only by the sequences that the traditional process could provide. Generating these synthetic de novo sequences is cheaper and faster as well, and could complement or even replace the traditional process exceptionally well.”
After antigen selection, the process of antibody discovery and development takes 2-3 years. Several breakthroughs have shortened this time frame, such as high-throughput screening of high-affinity binders, quick sorting of cells that make antibodies, parallel systems that assess multiple parameters of antibody candidates simultaneously and use of robotics to automate protocols. Phage display technology allows generating libraries of antibodies created in vitro (outside of a living organism). Such libraries provide a versatile platform for creating and selecting antibodies with tailored properties.
But even when the process is supplemented by faster, more advanced methods, developability remains an issue. When the extracted antibodies are validated in the laboratory, it turns out that the molecules need to be heavily modified for all the desired properties. Sometimes, the candidate loses its structure or function in this process.
This traditional route is not only lengthy but also suffers from a low success rate. Only 2% of candidates get through the checkpoints of preclinical and clinical testing. This late-stage failure of drugs usually means abandoning the program and starting over.
The Synthetic Route of Discovery
The solution lies in replacing the animal source with using Artificial Intelligence (AI) and Machine Learning (ML) to increase efficiency while reducing time and cost. “This synthetic route will drive the industry towards making antibody discovery cheaper, faster, and less risky,” Dr. Mathur adds.
CEO of Symphonytech and ex-Head of R&D at Biocon, Dr. Narendra Chirmule agrees. Talking about his several decades of journey in the field, he says, “What I've learned now is that AI has an exponential effect on predictability.” He elaborates, “using machine learning in the early process, combined with our understanding of the biology of the disease, we can design better screening processes and improve the probability of success.”
The giants of the pharma industry are actively collaborating with AI companies. Meanwhile new companies are entering the market with innovative approaches. AI-ML has been adopted in biologics only recently, while it has existed in the small molecule space for about 20 years.
Indian companies like Bharat Biotech and Serum Institute of India have proven their prowess in the global market during COVID. We are the largest manufacturers of generics. However, there is a need to shift the focus of Indian pharma from generics to novel drugs. “Our talent pool in biology, mathematics and computer science could act as a game changer with emerging technologies," says Dr. Chirmule.
Protein before and after folding (from Wikimedia Commons)
AI-based design of novel antibodies
Several companies are using AI-driven approaches to design novel antibodies. AI's footprint in the field is still in its early days, opening doors for innovation.
There are several databases which can be used to derive information on the sequence and structure of antibodies. Deep learning models can predict how proteins fold from sequences alone. AI is being employed to optimise new antibodies against known target antigens. Additionally, ML models can also be used to predict associated features of antibodies from sequence. However, these antibodies when generated from their designed sequence often fail to fold correctly in the lab. The 3D structure of the antibody determines how it interacts with an antigen. This antibody structure information is critical for its properties such as specificity and affinity.
Another common strategy involves predicting mutations in existing antibodies using AI to enhance their attributes like specificity and affinity, while preserving therapeutic function. However, these predicted mutations might improve one property while causing loss of another important one. This means going back and making modifications to the molecule, adding time and resources to the process of discovery. Thorough experimental validation remains indispensable in these methods. Sometimes, the validation reveals that the candidate does not possess all the properties to be developed into a viable drug. Thus, a purely sequence-based method is not the optimal solution to the challenges of antibody discovery.
immunitoAI Bridging the Gap
Despite an influx of inventive ways to use AI to generate novel antibodies, there are still unmet needs in the field. Only sequence-based design adds immense load on the later phases of drug discovery: the biological validation phase.
At immunitoAI, we use deep learning models to design novel antibodies, without a biological source. We believe that the sequence and structure of antibodies are crucial. We carefully consider both the sequence and structure of the molecule, along with its chemical interactions and binding properties. The neural networks learn all the necessary biological features of the sequence and mathematical representation of the structure to generate optimal designs of possible antibody candidates. Our pipeline computationally designs and ranks these antibody candidates. We then synthesise, characterise and validate these antibodies in our inhouse laboratory.
Our pipeline promises faster and more efficient drugs compared to the traditional method
We are using a drug-first approach, where the desired characteristics and therapeutic properties are also introduced in the designing phase of our AI pipeline, both at the sequential as well as structural level. This drastically reduces the iterative exploratory phase and cuts down on biological validation times. With our approach, the entire process from antibody design to experimental validation is expected to take less than a year. Since the drug properties are pre-engineered, the chances of success are very high at this stage.
“Synthetic drug discovery is the right thing to do. A lot of people are attempting it. I think what differentiates them (immunitoAI) is the ‘how’ part. To me, it's an inflection point in the antibody discovery space,” says Dr. Mathur. The unique approach of completely eliminating the biological source and focusing on engineering the antibodies as drugs, gives us an edge. With this efficient and faster approach towards antibody drug discovery, our aim is to create a sustainable model that makes antibody therapy as commonplace and affordable as everyday medicines.