From Receptor Signaling to Therapeutic Design: Rethinking Antibody Discovery
By Anabella Villalobos, Ph.D., Chief Scientific Officer, Metaphore Biotechnologies

Ninety-five percent of today's therapeutic antibodies work by blocking biology, leaving an enormous opportunity for biologics that can do more. Many diseases require pathways to be activated. Others require signals to be tuned, multiple targets to be engaged, or several functions to work together in one simple medicine. These are therapeutic challenges that blocking alone cannot solve. We connected with Anabella Villalobos, Ph.D., CSO of Metaphore Biotechnologies, to dive deeper into this.
Metaphore says it measures biological activity in living cells to design novel antibodies. What specific types of functional data are most valuable to the platform, and how do those data change the way you select or design an antibody candidate?
The differentiating value is to bring functional data into antibody design from the beginning, alongside binding and the properties a molecule needs to advance as a medicine. The most useful data show what an interaction causes a cell or biological system to do. Receptor activation, downstream signaling, and selectivity help us assess whether an antibody can activate a target, favor a particular pathway, or distinguish between related targets.
Through functional deep mutational scanning, we aim to systematically vary individual amino acids in a protein sequence and measure how those changes affect biological activity in living systems. For a receptor of interest, the intention is to mutate every amino acid in that target to every other possible amino acid. We express the variants at library scale in mammalian cells and capture activation of specific pathways using fluorescence or transcriptional readouts. Coupled to library-scale functional selections, this helps identify molecular features that preserve, enhance, or reduce a response. Our computational models are designed to learn from these data sets to identify molecular features that can guide antibody design and selection. This will enable us to design antibodies that can control complex biology in ways that extend far beyond simply binding and blocking a target.
Among antibody actions such as agonism, biased signaling, multitarget activity, and selective antagonism, which are most challenging to engineer, and what have you learned about making them more predictable?
There isn’t just one answer. It depends on the target, the response needed, and the properties required for the final molecule. Agonism and biased signaling can require a very specific mode of engagement. Multitargeting activity adds the challenge of making individual activities work together in one therapeutic molecule. Selective antagonism requires distinguishing the interaction we want to block from related biology we want to preserve.
What we have learned is that binding alone does not tell us whether an antibody will deliver the desired activity. However, cAMP signaling can provide a functional readout of GPCR activation. Concentration-response measurements help characterize both potency and the maximum response an antibody can produce. To assess biased signaling, responses across different pathways, such as cAMP signaling and β-arrestin recruitment, can be compared against a reference agonist. Selectivity requires evaluating activity at the intended target relative to related targets.
We then use this functional data set to train our machine learning models, which become more effective at designing, predicting, and increasing the number of antibody-based agonists. This is powerful because it allows us to pursue targets with precision.
How do you distinguish a useful predictive association from a causal explanation of a biological response in large functional data sets?
A predictive association tells us that a molecular feature is linked to a biological response. High-throughput models (such as generative sequence-to-function models) can successfully leverage predictive associations as proxies for activity. We use these model predictions to identify designs worth testing.
Establishing causation requires experiments designed to test if a specific biological activity is retained or lost when the predicted interaction is changed. The question is whether changing a predicted feature alters the response in the way the model suggests. Interpreting that result also requires considering other explanations, such as differences in protein expression or binding, that could affect the measured activity. Though we leverage computational models that help us continuously improve predictions about a system or an agonist we are designing, it is even more important to rigorously and robustly generate the experimental data that proves whether we are right.
What does the cycle from target selection through functional experiments, computational design, and validation look like experimentally?
We start by defining the response a medicine needs to produce and building assays that can measure it. In living systems, we test how diverse molecular changes affect that response. Our computational models learn from the resulting data and propose new agonists for further evaluation.
We then make the new designs and assess their activity, specificity, and developability. These dimensions need to be considered together. An antibody with strong activity still needs appropriate selectivity and properties that support its production and further development. For example, we measure specific developability properties — such as expression (titer), aggregation (SEC, DLS), hydrophobicity (HIC), self-association (AC-SINS), and thermostability (NanoDSF) — for every antibody we express.
Those results continue to train our models and inform the subsequent rounds of design. The assays and development criteria vary by program, but the connection between experiment, prediction, and testing remains central to the process.
How do you determine whether an antibody response observed in a cellular assay will translate into an animal model and ultimately a human therapeutic setting?
At the outset of a program, we define the relevant species cross-reactivity requirements and build them into molecule design, functional screening, and cellular readouts as much as possible. Following high-throughput cellular results, we will test our promising lead molecules in relevant in vitro pharmacology assays and in vivo disease models, measure exposure and establish pharmacokinetic/pharmacodynamic (PK/PD) relationships, and profile for developability to continue further development. The aim is to understand whether exposure to the antibody produces the intended biological or therapeutic effect, how that effect changes with dose, and how long it lasts. Those findings help connect cellular activity with a therapeutic outcome and potential dosing approach.
We also leverage state-of-the-art computational modeling, including physiologically based pharmacokinetic (PBPK) modeling, and related computational approaches to predict how our candidate drugs may behave in humans (e.g., doses and dosing frequency, tissue exposures) and use those projections to inform development decisions.
The assays and models depend on the target and disease. Cellular and animal studies, including safety assessment, help us decide whether a candidate warrants further evaluation, but only clinical studies can firmly establish whether it delivers the desired benefit in people.
How do you decide which biological problems are particularly well suited to Metaphore’s approach in metabolic disease and immunology?
In our world, the disease problem comes first. We look for a meaningful patient need, strong biological rationale, clinical doability, commercial opportunity, and a clear match to our platform. Metabolic diseases and immunology include opportunities where treatment may precisely require receptor activation, selective control of signaling, or activity across multiple targets at once.
We also ask whether our platform approach can address the problem experimentally: Can we build relevant assays, discover new mechanistic insights that lead to a differentiated solution, evaluate the biology in appropriate models, achieve the desired durability of therapeutic effect, and define a clear development path with biomarkers? Just as importantly, does connecting functional experimentation with computational design give us an advantage over other approaches in designing an antibody with the biological activity needed to address it?
We make disciplined choices where evidence supports a clear scientific and development rationale.
What would constitute a convincing demonstration that this approach represents a meaningful advance in antibody discovery, rather than simply another way to conduct antibody engineering? What milestones should the industry watch for?
Metaphore is positioned in the design of antibodies that activate and tune complex biology, not just block it. This represents a meaningful differentiation and advance in antibody discovery. Our goal is to establish that we can repeatedly deliver antibodies with useful biological activity and the properties needed for development across multiple programs. One successful molecule is an important first step, but it cannot establish the breadth of a platform on its own.
The industry should ultimately look for us to generate new antibodies with desired biological activity (antibodies that do what biology needs them to do, not just block) and clinical results that demonstrate a meaningful benefit for patients. In the meantime, we are focused on demonstrating the power of functional data and the delivery of development candidates with a clear therapeutic rationale.
The standard should be high, and we are building that evidence through our pipeline. We are excited to share our progress through scientific presentations, publications, and clinical development announcements.
About The Expert
Anabella Villalobos, Ph.D., joined Metaphore as chief scientific officer in 2025, bringing more than 30 years of experience in drug discovery and development. Previously, she served as senior vice president and head of biotherapeutics and medicinal sciences at Biogen. She led the delivery of differentiated drug candidates for neurological, rare, and autoimmune diseases across small molecules, biologics, oligonucleotides, and gene therapy. Before Biogen, Anabella spent 28 years at Pfizer in roles of increasing responsibility, most recently as vice president of medicinal synthesis technologies and neuroscience medicinal chemistry. Her group contributed to the discovery of Ogsiveo® (nirogacestat), Juvmo® (tavapadon), and other first-in-class clinical candidates.