End-to-end automated machine learning at scale is the fastest route to tomorrow's most impactful antibody therapeutics.
End-to-end automated machine learning at scale is the fastest route to tomorrow's most impactful antibody therapeutics.
Antibody therapeutics must satisfy a wide array of biophysical conditions to be effective tailored therapies, but the search space is extremely large and impossible to search exhaustively. Iteratively designing new antibodies, characterizing their properties, and learning from this new data is the most effective way to create innovative new therapies. BigHat has built Reccy Antibody Design Studio to carry out this mission at unprecedented scale.
RADS offers a framework for optimizing any antibody, for any therapeutic property, for any indication. Each week, it trains a competitive zoo of models on the latest data, chooses the best one, and uses that model to design the next week’s antibodies. RADS offers a plug-and-play API to integrate any model or assay into its active learning infrastructure so that it is always leveraging the best model for the job. RADS natively integrates with our automated wet-lab and data processing facilities to move proposed antibodies rapidly to clinical candidates. This quick, iterative, and competitive process leaves no value on the table, ensuring that BigHat produces the best therapies as efficiently as possible.
Antibody therapeutics are among the world’s greatest opportunities to impact human health, with one of the biggest opportunities to unlock new treatments using next generation engineering. BigHat's Reccy Antibody Design Studio (RADS) is our fully integrated AI platform that orchestrates the thousands of models and datasets in continual use for our programs. RADS unifies machine learning with Milliner™, our high-throughput, automated wet-lab characterization platform to bring antibodies to patients faster through automated design-build-test cycles.
Antibody therapies are among some of the most safe and effective medicines for our most intractable diseases, from inflammation to infections to cancers.
Yet, we’ve only scratched the surface of what biologic drugs can do in fighting disease because clinical-grade antibodies also come with challenges – they must have high affinity and specificity (i.e. the probability of a drug binding occupying a receptor), low immunogenicity (i.e. must not provoke stimulate an immune response), high stability, and manufacturability. There are layers upon layers of dependencies when developing an antibody therapy.
Conventional antibodies can target and neutralize the disease-causing cells with few, if any, modifications, and have been used effectively to treat a variety of conditions. But they are limited in their abilities because they only bind to a single target. Next-generation antibodies can engage multiple targets simultaneously, but this approach often creates a “Frankenstein molecule” that can severely compromise the properties of the antibody – they often struggle with sufficient biophysical quality for therapeutic use.
There are now over 200 approved antibody therapeutics and nearly 1,400 in clinical development, fueling a global biologics market projected to reach $783 billion by 2030. Antibody drugs such as adalimumab (Humira™) and trastuzumab (Herceptin™) have already revolutionized the treatment of chronic inflammatory diseases and cancers, while next-generation formats like bispecifics and antibody–drug conjugates are expanding what’s possible in patient care.
BigHat developed the Milliner™ platform to overcome these challenges and create better antibodies faster and undertake novel designs far beyond what’s possible today.
Milliner integrates a synthetic biology-based high-speed wet lab with state-of-the-art machine learning technologies into a full-stack antibody discovery and engineering platform.
For antigens with a known structure, BigHat’s de novo antibody design models to generate epitope-specific hits fully in silico. Each antibody is ultimately fed into our iterative engineering platform for further biophysical and functional optimization.
Every therapeutic program begins with "seed" antibodies discovered by our in-house computational or experimental library display capabilities, provided by partners from previous discovery efforts, or identified in the public domain.
Standard assays consist of purity, yield, affinity, stability, solubility, specificity, and cell-based functionality. All sequences are designed to minimize potential immunogenic and developability liabilities. We leverage a suite of automated, functional disease-surrogate assays to meet the demands of our own internal and partnered therapeutic programs.
Hundreds of recombinant antibodies can be synthesized, purified, and fully characterized for biophysics and function in a single, weekly workcell. BigHat leverages state-of-the-art synthetic biology technologies such as DNA synthesis, cell-free protein synthesis, and scalable purification to quickly produce just enough protein for downstream assays. Every antibody is thoroughly characterized for biophysics and function at BigHat's headquarters.
At BigHat, every therapeutic program starts with a design blueprint and antibodies generated in our discovery engine or supplied by a partner. These initial molecules are iteratively improved through sequential design-build-test cycles. Our proprietary pre-trained machine learning models design hundreds of variants that we build and test in our lab each cycle.
We measure biophysical properties and impact on disease activity for every variant using cell-based or other functional assays that replicate in vivo disease processes. We then incorporate this new information into our models, iteratively accelerating our predictions. Over multiple cycles, our models quickly identify antibodies that match our design blueprint.
Design models balance sequence exploration and exploitation for smart antibody design
BigHat has been exploring the Milliner platform’s capabilities since our founding in 2019. Characterizing tens of thousands of antibodies over the course of hundreds of design rounds has taught us a lot.
The main lesson we learned is that Milliner's special blend of low-N active learning / ML technology and synbio-based rapid build/test wet lab is capable of solving some of today’s most challenging antibody design problems.
Reliably discover variants with better binding
Designer scaffolds improve an entire class of therapeutics
Iteratively improve affinity, stability, & function, together
Steer towards the true targets, away from tox-causing isoforms
Optimize proxies of disease biology ensuring therapeutic activity
Leverage microenvironments to augment safety and efficacy
Readily swap formats preserving biophysical optimization
ML-designed libraries and de novo design unlock & accelerate pipeline and partnerships
BigHat’s technology is widely applicable across biotherapeutic formats. Although we have focused on antibodies first, BigHat can optimize virtually any property of any biological molecule.
Today, BigHat specializes in next-generation antibody formats, including camelid and human VHHs, scFvs, multispecifics, and conjugates.
Differentiated therapeutics optimized for affinity, CMC, and function
Optimize antibody conjugates for end-to-end function
Reusable framework molecules unlock novel targets and capabilities
Augment the therapeutic properties of enzymes, hormones, peptides
Overseeing all aspects of the Milliner platform is BigHat’s custom-built, fit-for-purpose LIMS and operating system, Reccy. It controls instruments, coordinates robots, schedules people, generates orders, manages data, trains ML models, and everything in between. Reccy enables BigHat to run a smarter, quicker lab by making data and process best practices available to everyone, all the time.
A home-grown, cloud-native application, Reccy is designed with best practices for software engineering and cutting-edge security to guarantee data quality, integrity, safety, and access control. BigHat maintains a SOC 2 Type 2 certification, having passed our first audit with flying colors in 2022.
