Research program
Cracking the antibody code
Benjamin Nemoz, M.D., Ph.D. · antibody engineering · structural immunology · machine learning
One question runs underneath almost everything I do: what are the hidden rules that govern how a good antibody is made? How does a B cell start from an unremarkable germline sequence and, through somatic hypermutation and selection, arrive at an antibody that neutralizes a virus the immune system has struggled with for years? If those rules can be read, they can be learned; and if they can be learned, they can be used to design antibodies and vaccines on purpose.
My program pursues that question on three fronts at once: at the bench (antibody isolation, structure, and repertoire sequencing), in the clinic and the field (the pathogens that actually harm people, especially the neglected ones), and in code (foundation models of antibodies and the data infrastructure they need). It is deliberately aimed at targets the market will not fund, and it runs across a network of collaborators on four continents. What follows is an honest map, including what is finished, what is in flight, and what is still in the boxes.
This work is carried out at Scripps Research in the laboratories of Dennis Burton and Bryan Briney, without whom none of it would be possible.
Status ✓ done ◐ drafted ◆ in press ● current ▪ in the boxes ○ exploratory
Theme I
Reading the code
Can the rules of affinity maturation be learned from sequence alone?
The core of the program. If antibody maturation follows learnable rules, a model trained on enough antibody sequence should recover them, and then generate maturation that nature has not gotten around to yet. This is where the bench and the model close a loop.
Design, training, and evaluation of a foundation model of antibodies operating at nucleotide resolution, the level at which somatic hypermutation, codon context, and AID targeting actually act, and which amino-acid models cannot see.
A two-model system (mutation targeting + substitution) that simulates SHM from sequence context alone, with no antigen information in training. Applied to the broadly neutralizing HIV lineage PC94-A, predicted mutations improved neutralization potency up to 76-fold while preserving breadth.
The sugar layer as part of the code: N-linked glycosylation and its role in antibody function and B-cell immunobiology.
Theme II
The PC94-A lineage, end to end
From one elite-neutralizer donor, how deeply can a single broadly neutralizing lineage be understood, and pushed?
From the HIV-1 elite-neutralizer donor PC94, I isolated several antibody lineages using complementary methods, then characterized the broadly neutralizing lineage PC94-A (a V3-glycan / N332-supersite bnAb) end to end. Much of it was done hands-on across the whole stack, from operating the sequencer and producing the pseudoviruses for neutralization assays to the structural and biophysical work, and contributing fixes to the repertoire-analysis toolkit (abstar) that ties it together.
Recovery of PC94-A, among other lineages, from donor PC94, and end-to-end characterization of PC94-A against HIV-1 Env: neutralization assays, biolayer interferometry (BLI), cryo-EM, ELISA, and polyreactivity.
Using the SHM model to mature PC94-A further than it went in vivo; model-predicted variants preserved heterologous breadth with selective potency gains (up to 76-fold), validated in vitro.
Exploring whether constant-region (CH1) changes in PC94-A carry functional benefit, a direct probe at the antibody code beyond the variable domain.
What the cell biology looks like behind a lineage that reaches breadth (links to Theme I).
Theme III
Antibodies against neglected & hard targets
Can we reach the pathogens no one else will pay to reach?
The point of the program. Structure plus neutralizing antibodies, aimed squarely at high-consequence and neglected pathogens, the ones that never clear a commercial bar.
Broadly cross-reactive antibodies spanning Ebola Sudan, Marburg, and Bundibugyo.
Antibody discovery against Lassa, one of West Africa's most pressing hemorrhagic-fever threats.
Bringing patient samples to Scripps to solve the surface-glycoprotein structure and isolate antibodies.
Antibody design against Andes orthohantavirus; recent lecturing on hantavirus biology.
Solve the envelope glycoprotein structure and isolate neutralizing antibodies. Parked for now; a prospective sample-collection collaboration in French Guiana may restart it.
A long-standing interest and my first research love: dengue, chikungunya, and Zika (among others) at CHU Grenoble, including work on tick-borne encephalitis.
Inducing indels in influenza antibodies ◐ drafted, with Nitin Bhalchandra Mishra (postdoctoral associate, Wilson lab, Scripps); and in-silico-guided design of flu stem antibodies ● current, with Mahdi Shafiei (Ph.D. candidate, Scripps).
Theme IV
Vaccines — teaching immunity in advance
If we know what a good response looks like, can we elicit it on purpose?
The design questions run in both directions: not only reading antibodies, but engineering the immunogens meant to raise them.
Design of the PHI133 Env immunogen for an HIV mRNA vaccine candidate.
Design and evaluation of an mRNA vaccine based on PC39 founder-virus sequences (a second Protocol C donor).
Across these, I have built end-to-end mRNA-vaccine design and evaluation into the toolkit.
Theme V
Seeing the whole repertoire — data & infrastructure
What does it take to observe antibodies at the scale a model needs?
None of the modeling works without data at the right scale and quality. A large part of the program is the wet-lab and computational machinery to generate, pair, annotate, and curate antibody repertoires, including the training corpora the foundation models depend on.
A platform (wet lab and code) that recovers paired heavy/light chains at the scale of millions of pairs via combinatorial barcoding and multilayer demultiplexing.
Adding an antigen-barcoding layer so pairing carries specificity information. Work in progress.
Exploring heavy/light pairing across ~4.8M pairs, with molecular-dynamics simulations over ~13,000 representative pairings, aiming to release two of the most substantial paired-antibody datasets available.
Memory-B-cell-enriched bulk sequencing for repertoire-scale studies and for building the training corpora the antibody foundation models require.
Constructing donor-specific germline references, from RNA (IgDiscover) and from DNA (custom pipeline, Ig-locus sequencing on Nanopore). Beginning with Californian donors, expanding to Kenyan donors.
Antibody-repertoire work in African donors, extending repertoire and germline science beyond the usual populations.
Theme VI
Prediction, methods & benchmarks
Can we predict antibody behavior, and prove it in the open?
Modeling claims only matter if they hold up prospectively. This strand is about predictive methods and about testing them in blinded, wet-lab-validated competitions.
Training models to predict antibody–antigen binding affinity, including exploration of quantum-computing approaches alongside classical methods.
A blinded, prospective benchmark of in-silico antibody discovery (Nature Biotechnology). A team effort with the Briney lab and Nitesh Mishra (staff scientist, Scripps); our submissions performed strongly across the challenge tasks.
Prospective antibody-design competition against Nipah, another neglected, high-consequence paramyxovirus.
Theme VII
Translation — from model to molecule
Can this actually make medicines?
The program is built to produce real molecules, not only papers.
Co-founder & CTO, with co-founders Felix and Stephen. Semi-autonomous design of next-generation, fully human therapeutic monoclonal antibodies. Bootstrapped; French national “Deep Tech” label (Bpifrance). Programs include Alzheimer's nanobody design (with PepperPrint, Germany), alongside antibody-design work for external partners.
Antibody-design programs with DKFZ (German Cancer Research Center) and a NASDAQ-listed biotech; nanobody work with PepperPrint (Germany).
Reach
The collaboration network
This is, by design, a distributed program, built on the work of many people. At its center, Dennis Burton and Bryan Briney at Scripps Research, with colleagues Karenna Ng, Mahdi Shafiei, Nate Price, Nathan Liendo, Nitesh Mishra, and Jonathan Hurtado; structural biology in the Andrew Ward lab (Gabriel Ozorowski) and influenza engineering with Nitin Bhalchandra Mishra (Wilson lab); glycobiology with Max Crispin and Joel Allen (University of Southampton); molecular dynamics with Monica Fernández-Quintero. Clinically and internationally, with Julien Lupo and Giovanna Clavarino (CHU Grenoble Alpes), Nicolas Ellinger (Lyon), Facundo Batista (Ragon Institute, MGH/MIT/Harvard), Eunice Nduati (KEMRI-Wellcome Trust, Kenya), and a prospective effort in French Guiana. In industry: Fabulous Sciences (with co-founders Felix and Stephen), DKFZ, PepperPrint, and a NASDAQ-listed biotech partner.
Where this is going
An independent research program
Pulled together, these threads describe a single, fundable program built around a closed loop: sequence → model → designed antibody or immunogen → bench validation → back into the model. Three pillars carry it.
1. The antibody code. Nucleotide-level foundation models of antibodies and somatic hypermutation, extended with glycosylation, constant-region, and cell-state biology: a mechanistic account of how affinity and breadth are actually built, and a generative engine that can extend maturation beyond what nature has sampled.
2. Neglected & hard targets. Turning that engine on the pathogens the market ignores, filoviruses, Lassa, hantaviruses, HTLV-1, arboviruses and Oropouche, Nipah, influenza, pairing structural biology with model-guided antibody and vaccine design.
3. The discovery engine. The data and methods that make the loop run at scale: natively-paired repertoire sequencing with specificity, globally representative germline references (California to Kenya), and prospectively validated predictive models.
The near-term ambition is to make that loop routine, and to point it first at the diseases with the fewest advocates. Antibodies are one of the few therapeutic modalities where design is genuinely within reach; the goal of this program is to make principled antibody design fast, general, and aimed where it is needed most.