Unlike conventional drugs, which behave the same way everywhere, these programmable antibodies can sense the micro-environment in a specific patient at a specific time and respond dynamically.
Novel Capabilities
Precision binding to preselected epitopes
By designing antibodies that bind to predefined epitopes — the parts of antigens recognized by antibodies — we can create drugs that achieve specific functional effects. These antibodies can thus agonize or antagonize the target by precisely “pressing the right button”, they block one function of the target while sparing, or even enhancing, another function and give us better control of the biological process.
This enables Biolojic’s antibodies to have better efficacy, broader therapeutic windows and better dosing regimens.
Example – Aulos: Selective effector cells agonist for IO
This is illustrated by AU-007, the first-ever computationally designed antibody to enter human clinical trials. AU-007 was designed by Biolojic to precisely bind to pre-defined epitope on interleukin-2 (IL-2) to redirect IL-2 toward cells that express dimeric receptors (T effs, NKs, NKTs), and away from cells that express the trimeric receptor (T regs, vascular endothelium). Thus, AU-007 unlocks the full potential of IL-2 to fight cancer: it starves the cells that protect the tumor and nourishes that cells that kill the tumor.
Interleukin-2 is often described as a “double-edged sword” because of its ability both to suppress and activate the immune system. For cancer patients, increased levels of IL-2 correlate with improved survival.
A multibody is an antibody that can bind to two or more targets on each of its arms. It is a standard symmetrical IgG antibody, in which each arm of the antibody is multi-specific. This structure allows us to attack different targets with a single drug. A multibody can act like different antibodies at different times, depending on the conditions in the body. They are essentially programmable nanobots that can execute “and” or “or” functions inside the body.
For example, we are developing a multibody that can interact with cancerous tumors as well as chemotherapy. We have designed the antibody to grab chemotherapy in the body and then bind to a tumor and deliver the chemotherapy. In this way, the antibody essentially mops up chemotherapy from places where it could do damage and concentrates it where it can do the most good — in the tumor microenvironment.
Designing Programmable Antibodies
When your immune system encounters an antigen, it searches for an existing antibody (in the immune memory or in the germline) that will serve as a template to create a new antibody for that specific antigen. Next, your immune system begins making specific mutations to that template antibody, creating a new antibody that will bind to the new target.
We take a similar approach, using artificial intelligence (AI). Over the past decade, we’ve collected billions of measurements on millions of antibodies. This rich proprietary dataset informs computer algorithms that help design antibodies with new capabilities. In each project, we generate millions of new data points that allow us to train and refine dedicated machine learning models specifically for the task at hand.
- Input – We start with the desired target product profile (TPP), which defines the characteristics we want the antibody to have — for example, what target or targets it binds as well as how it binds.
- Our first AI model identifies a template antibody that can serve as the starting point and suggests specific mutations that will cause the selected template antibodies to bind the desired target(s) with the desired functional effect
- High throughput experiments generate data on the effect of specific mutations on antibody function and characters
- Our second AI model is trained on the new experimental data
- Output – A sequence that is predicted to meet the TPP, which can be characterized and validated experimentally
Fueled by Big Data
Just as we continually refine our algorithms, we continually grow our data. This data is the foundation for our platform — the information that is used to train our machine-learning models. It includes proprietary experimental high throughput data we’ve gathered through billions of experimental measurements spanning more than a decade of antibody engineering work, as well as structural databases of all known antigen-antibody pairs and vast data sets of human antibody repertoires. As we design new antibodies and conduct new experiments, we gather new data, which strengthens the platform.
For each project, we generate in the lab millions of additional experimental measurements using surface display techniques combined with deep sequencing, on sequences that were designed specifically for this project. The data that we collect in these project-specific experiments allows us to train and refine dedicated models for each project.