Next Phase Newsletter

Juan Carlos Lopez

Andy Marshall

Although therapeutic antibodies represent a $160 billion-dollar annual market and comprise a third of all approved drugs, discovering new antibody molecules remains a labor-intensive process, requiring slow experimental approaches with low hit rates, such as animal immunizations and or the panning of phage- or yeast-displayed antibody libraries. The drug hunter’s dream would be to design an antibody to any target by simply entering information about that epitope into a computer. Now that dream is one step closer with a recent proof of principle peer-reviewed paper published in Nature on work disclosed last year from the team of 2024 Nobel Laureate David Baker. Baker and his colleagues at the University of Washington introduce the first generalizable machine-learning method for designing epitope-specific antibodies from scratch without relying on immunization, natural antibody repertoires, or knowledge of pre-existing binders.

Unlike small-molecule drug development, which has benefitted from an explosion of interest in the use of machine-learning models, in-silico design of antibody binders has lagged far behind. One reason for this is the paucity of high-resolution structures of human antibody–antigen pairs—currently only ~10,000 structures for 2,500 antibody-antigen pairs have been lodged in SAbDab (a subset of the RCSB Protein Data Bank). Most of these structures are soluble protein antigens, but there’s little data to model antibody binders to GPCRs, ion channels, multipass membrane proteins and glycan-rich targets, which are of most commercial interest. Overall, the antibody–antigen structural corpus is orders of magnitude smaller, noisier and narrower than that available for small molecules, lacking information on binding affinities and epitope competition maps via PDBBind/BindingDB/ChEMBL.

For these reasons, most companies have focused on machine learning prediction of developability properties—low aggregation, high thermostability, low non-specific binding, high solubility, low chemical liability/deamidation and low viscosity—for an antibody’s scaffold, rather than in-silico design of the six complementarity determining-regions (CDRs) on the end of an antibody’s two binding arms.

Even so, several recently founded startups have claimed to be using machine-learning models to predict/design antibody binders from scratch. These include Xaira Therapeutics, Nabla Bio, Chai Discovery and Aulos Bioscience.

Xaira debuted last year with >$1 billion in funding to advance models originating from the Baker lab. Nabla Bio also raised a $26 million series A in 2024, publishing preprints in 2024 and 2025 that describe its generative model (‘JAM’) for designing VHH antibodies with sub-nanomolar affinities against the G-protein coupled receptor (GPCR) chemokine CXC-motif receptor 7 (CXCR7), including several agonists. In August, Chai announced a $70 million series A financing based on its ‘Chai-2’ generative model disclosed in a preprint that details de novo antibodies/nanobodies against 52 protein targets, including platelet derived growth factor receptor (PDGFR), IL-7Rα, PD-L1, insulin receptor and tumor necrosis factor alpha, with “a 16% binding rate” and “at least one successful binder for 50% of targets”.

Finally, Aulos emerged with a $40 million series A in 2021 as a spinout from Biolojic Design. This program has generated computationally designed de novo CDR binders with picomolar affinities for epitopes on HER2, VEGF-A, and IL-2. The IL-2 antibody (imneskibart; AU-007)—designed to selectively bind the CD25-binding portion of IL-2, while still allowing IL-2 to bind the dimeric receptor on effector T cells and natural killer cells—reported positive phase 2 results in two types of cancer just last week. Absci, another more established company, has also been developing de novo antibodies, publishing a generative model for de novo antibody design of CDR3 loops against HER2, VEGF-A and SARS-CoV-2 S protein receptor binding domain.

Overall, though, computational efforts have largely optimized existing antibodies or proposed variants once a binder already exists. Recent generative approaches have often needed a starting binder, leaving de novo, epitope-specific antibody creation as an unmet goal. The Baker paper now provides a generalizable, open-source machine-learning approach that can find low nanomolar antibody binders to a wide range of targets.

To accomplish this task, the authors use RFdiffusion, a generative deep-learning framework for protein design, extending its capabilities by fine-tuning it specifically on antibody–antigen structures. Their goal was to enable the in-silico creation of heavy-chain variable domains (VHHs), single-chain variable fragments (scFvs), and full antibodies that target user-defined epitopes with atomic-level structural accuracy.

Their approach integrates three major components: backbone generation with a modified RFdiffusion model, CDR sequence design via the algorithm ProteinMPNN, and structural filtering using a fine-tuned RoseTTAFold2 predictor (the authors note that improved predictions can now be obtained by swapping out RoseTTAFold2 for AlphaFold3 developed last year by Google Deepmind and Isomorphic Labs). The refined RFdiffusion model can design new CDRs while preserving a fixed antibody framework and sampling diverse docking orientations around a target epitope. The resulting models generalize beyond training data, producing CDRs unlike any found in natural antibodies.

Baker and his colleagues created VHHs against several therapeutically relevant targets, including influenza H1 haemagglutinin, Clostridium difficile toxin B (TcdB), SARS-CoV-2 receptor-binding domain, and other viral or immune epitopes. High-throughput screening via yeast display or purified expression led to the identification of multiple binders, typically with initial low affinities in the tens to hundreds of nanomolar range. Cryo-EM confirmed near-perfect structural agreement between design models and experimental complexes, particularly for influenza haemagglutinin and TcdB, demonstrating atomic-level accuracy across the binding region and the designed CDR loops. To enhance affinity, the authors used OrthoRep, an in-vivo continuous evolution system, for the affinity maturation of selected VHHs. The affinity of the resulting VHHs improved by roughly two orders of magnitude while retaining the original binding orientation.

Baker and his team further challenged their method with the more difficult problem of de-novo scFv design, which requires simultaneous construction of six CDR loops across two amino acid chains. The team introduced a combinatorial assembly strategy in which heavy and light chains from structurally similar designs were mixed to overcome cases where a single imperfect CDR would compromise binding. This enabled the discovery of scFvs targeting the Frizzled epitope of TcdB and a PHOX2B peptide–MHC complex. Cryo-EM validation of two scFvs showed that all six CDR loops matched the design model with near-atomic precision.

Future work is needed to extend de novo antibody prediction via this method to tougher target classes, such as membrane proteins. Clearly, modeling across all six CDR loops and the heavy and light chains remains a hard problem; indeed, the paper’s marquee result was designing a single scFv where all six CDRs matched the designed pose at high resolution; more generally, scaling reliable heavy- and light-chain co-design beyond a few cases remains an open engineering challenge that future methods will need to solve. For the field to gather momentum, benchmarking efforts like the AIntibody challenge will be needed, together with public efforts to create datasets of negative binding data, akin to those described in a paper published earlier this year.

Overall, the Baker paper is seminal work that establishes a practical and accurate approach to designing epitope-specific antibodies from scratch. It represents a major advance in the development of therapeutic antibody discovery.

Tags: AI, alzheimers, art, artificial-intelligence, beauty, biology, books, cancer, fitness, food, genetics, health, huntingtons-disease, immune-system, immunotherapy, nutrition, painting, poetry, science, technology, wellness, writing

Juan Carlos Lopez

Andy Marshall

This year’s Nobel Prize for Physiology or Medicine was awarded to Mary Brunkow, Fred Ramsdell and Shimon Sakaguchi for the discovery of regulatory T cells (Tregs)— white blood cells whose role it is to suppress overactivation of our immune system. The prize was unusual in that Brunkow made her discoveries while leading an industry R&D team at Darwin Molecular (now defunct). Ramsdell and Sakaguchi are also co-founders of two prominent biotech companies developing Treg therapies: Ramsdell’s Sonoma Biotherapeutics is developing autologous Treg therapies against arthritis and hidradenitis suppurativa, together with a LFA3-IgG1 fusion molecule for depleting CD2+ effector T cells; and Sakaguchi’s Coya Therapeutics is developing a low-dose interleukin 2 (IL-2)/CTLA-IgG1 fusion combination for amyotrophic lateral sclerosis and other neurodegenerative disorders; the Nobel prize likely helped boost Coya’s announcement in October to raise $20 million in follow-on funding on the public markets.

Tregs have long attracted the attention of drug developers interested in autoimmune conditions, diseases where the immune system is overactive. But progress in this field has been slow, and the first clinical results for T-reg cell therapies are only now beginning to emerge in liver transplantation and kidney transplantation. (Low-dose IL-2 treatments that promote Tregs have also begun to show promise in lupus and systemic sclerosis patients.)

The overarching idea behind Treg cell therapy has been to isolate these cells from a patient, introduce/upregulate expression of the FOXP3 transcription factor that marks them from other T cells, and expand them before giving them back to the patient.

Early attempts to develop this autologous therapy failed in part because Tregs are less numerous in the peripheral blood than effector CD4/CD8 T cells, difficult to isolate and problematic to expand. Moreover, the isolated Tregs are polyclonal, targeting multiple antigens. Approaches that expanded this unmodified polyclonal population of cells and put them back into patients resulted in a ‘diluted’, clinically insignificant, therapeutic effect.

To address this problem, companies are now turning to leverage advances in the chimeric antigen receptor (CAR)-T cell therapy field. A whole slew of Treg cell therapies is being engineered with CARs or T-cell receptors (TCRs), allowing targeting to specific antigens in specific organs.

As we mentioned above, the most advanced of these are in the organ-transplantation field, where chronic immunosuppression renders patients susceptible to infections that can be lethal. Sangamo Therapeutics’ TX200 and Quell Therapeutics’ QEL-001 are CAR-Treg therapies for renal- and liver-transplant rejection, respectively. These assets, which are in phase 1/2, both bind to human leukocyte antigen HLA-A2, which is exclusively expressed on the transplanted donor organ, ensuring that the Tregs travel exclusively to the place where they are needed. Elsewhere, Sonoma is also developing an autologous CAR-Treg therapy, SBT-77-7101, that targets citrullinated proteins abundant in rheumatoid arthritis (for which Sonoma recently announced positive interim phase 1 data) and the skin condition hidradenitis suppurativa.

A second focus for companies has been on TCR-engineered Tregs. The great theoretical advantages of TCRs over CARs are that 1) they have high sensitivity at low antigen density, 2) they focus exclusively on antigen-presenting cells which then reeducate/suppress effector T cells; 3) they don’t bind soluble antigen and 4) most autoimmune diseases are driven by intracellular proteins presented as processed peptides in the context of HLA. As yet, however, only a few companies are pursuing the approach. One example is GentiBio, which is developing GNTI-122 for type 1 diabetes. This Treg product expresses a TCR targeting a fragment (IGRP 305–324) of the pancreatic islet-specific antigen glucose-6-phosphatase catalytic subunit-related protein (IGRP). Another pioneer in this area, Abata Therapeutics, had also been developing a TCR-engineered Treg therapy (targeting myelin peptide/HLA-DRB1*15:01 for multiple sclerosis); however, the frosty financing environment in the first half of 2025 meant it ran out of cash and Abata closed its doors in August.

One challenge that all Treg cell therapies face is the plasticity of these cells and their tendency to shape shift into effector T cells, a phenotypic change that, in the therapeutic setting, could lower efficacy or even exacerbate pathology. One approach to address this problem has been to modify the cells by overexpressing the transcription factor FOXP3, the master regulator of Treg development. For example, as methylation of the FOXP3 promoter under inflammatory conditions can turn Tregs Into effector T cells, Quell’s Tregs are engineered with a methylation-resistant FOXP3 that compels the cells to remain in their suppressor phenotype. And to bring us back to where we started, Nobel laureate Sakaguchi turns out to be a serial entrepreneur, founding another company, Regcell, that recently relocated from Japan to the US on the back of a $45.8 million financing back in March. The company is using small-molecule CDK8/19 inhibitors that act as epigenetic modulators to lock in FOXP3+ Tregs that show a stable suppressive phenotype in vivo.

But Treg cell therapies still face stiff competition. Ironically, perhaps, from their antithesis: the effector CAR-T cell. Pioneering work by Georg Schett’s group at Friedrich Alexander University Erlangen-Nuremberg has galvanized numerous efforts to develop CAR-T depleters of pathogenic B-cell or plasma-cell subsets in autoimmune conditions. Evidence is growing for the clinical efficacy of this approach in diseases such as lupus or myasthenia gravis.

But the holy grail would be to dispense with cell therapy altogether and promote Treg activity in situ, without the need for purification and modification/expansion outside the body. By focusing on injectable biologics, many companies can bring products to market that are easily accommodated into current clinical practice, dispensing with the need for leukopheresis (an approach alien to most rheumatologists) and the complex logistics of ex vivo cell therapy.

​Nektar Therapeutics’ rezpegaldesleukin is a pegylated IL-2 given at low doses that acts on CD25, the high-affinity IL-2 receptor enriched in Tregs. The company recently reported positive phase 2 data in atopic dermatitis. Elsewhere, Egle Therapeutics and Mozart Therapeutics have discovery programs developing bispecific antibody Treg engagers for multiple autoimmune diseases. TrexBio has developed a peptide agonist of tumor necrosis factor receptor 2 (TNFR2), announcing in June the dosing of its first participant in a phase 1 trial for atopic dermatitis and other inflammatory diseases. Zag Bio is another T-cell engager play that recently came out of stealth,

The Treg field can rightly celebrate its Nobel recognition and the progress made towards bringing this cell type to patients. Although it will likely be several years before we gain a full picture of how Treg biology can be leveraged to fight autoimmune disease, the field eagerly awaits the readout from early efficacy trials of cell therapies and potentially an FDA-approved product for the biologics in later development.

Tags: AI, alzheimers, art, artificial-intelligence, beauty, biology, books, cancer, fitness, food, genetics, health, huntingtons-disease, immune-system, immunotherapy, nutrition, painting, poetry, science, technology, wellness, writing

Juan Carlos Lopez

Andy Marshall

On September 24, uniQure reported 36-months positive topline data from the phase1/2 study of their candidate AMT-130 for the treatment of Huntington’s disease. AMT-130 consists of viral vector AAV5 and a synthetic miRNA that targets exon 1 of the huntingtin gene. The results showed that AMT-130, directly injected into the striatum at a dose of 6 x 10^{^13} genome copies per subject, slowed disease progression at 36 months, as measured by the composite Unified Huntington’s Disease Rating Scale and by Total Functional Capacity compared with a “propensity score-matched external control”.

The results have yet to appear in the peer-reviewed literature, and some experts have urged caution in their interpretation, particularly with regard to the use of external historical control groups and the small number of patients (12 have completed the 36-month period). However, uniQure’s data have been widely welcomed as a breakthrough for a field that has experienced its fair share of false starts (most recently Roche/Ionis halting of its phase 3 dosing of tominersen in 2021 after promising phase 1/2a results). Moreover, the findings have bolstered interest in therapeutic approaches targeting exon 1 in the mutant allele in addition to reducing levels of the full-length huntingtin protein.

Huntington’s disease is a triplet repeat disease in which the huntingtin gene’s exon 1 bears the CAG repeat encoding the polyglutamine stretch that defines the pathology. It’s therefore not surprising that the N-terminal part of HTT and its product have attracted attention as drug targets. Broadly speaking, scientists have tried to get at exon 1 in three ways: targeting the gene itself to block transcription, targeting the mutant mRNA to inhibit translation, and targeting the truncated protein that results from the mutant mRNA. A recent review provides a thorough survey of the preclinical work on these three fronts.

From the drug-discovery point of view, the most advanced programs focus on the development of ASOs or RNAi sequences against the CAG repeat in the mutant mRNA. The motivation behind this strategy is in part the realization that transcription of mutant HTTexon 1 results in a shortened 102 nt mRNA that encodes a toxic protein prone to aggregation: HTTexon1.

To explain what goes wrong in RNA splicing, we need to take a quick detour into the biochemistry of mRNA processing. In any cell, pre-mRNA processing is a competition between the splicing machinery (which removes introns from transcribed genes by recognizing an intronic 5′ splice site, branch point, and 3′ splice site) and the machinery that carries out intronic polyadenylation. Intronic polyadenylation cleaves transcripts within introns and adds a poly(A) tail to the shortened exon–intron fragment transcript when intronic sequences like AAUAAA are present together with a downstream U/GU-rich element.

All of the above is important for Huntington’s because, in healthy brains (specifically the striatum), U1 small nuclear ribonucleoprotein (snRNP) is thought to sit on the cryptic polyA sites in intron 1 of HTT, blocking intronic polyadenylation and enabling accurate splicing of introns and production of a full-length (9,500 nt) mature HTT mRNA. In contrast, in Huntington’s patients, increasingly long CAG repeats in the huntingtin pre-mRNA are thought to sequester U1 snRNP, thereby interfering with formation of the spliceosome complex and making cryptic polyA sites accessible. The result is premature termination of transcription within intron 1, resulting in the generation of the the shortened 120 nt HTTexon1 mRNA transcript that encodes an N-terminal 17-amino acid HTTexon1 protein.

Until the UniQure program, most disease-modifying therapies in the clinic have sought to downregulate full-length huntingtin and haven’t discriminated between mutant protein and wild-type protein. The prevailing thinking has been that going after full-length HTT makes sense because both the full-length protein—and fragments of it produced by proteolytic degradation—were likely the main problem.

By targeting exon 1, AMT-130 aims to specifically reduce production of toxic HTTexon1. And several other drug developers have also started to pivot and focus more closely on targeting HTTexon1, with the hope that such approaches might have greater efficacy in reducing huntingtin aggregate nucleation.

Just this year, Alnylam/Regeneron recently took ALN-HTT02 into phase 1b testing. This siRNA is conjugated to a 2′-O-hexadecyl C16 palmitate lipid that enables traversal of the blood brain barrier. It targets a conserved mRNA sequence within huntingtin exon 1, leading to the RISC-mediated degradation of all HTT mRNAs. The approach downregulates both HTTexon1 and full-length HTT — and does not discriminate between the wildtype and mutant alleles.

There are other molecules in development that directly target the expanded CAG repeat in exon 1 that are allele-specific. Vico Therapeutics’ VO659 is an ASO with an allele-preferential mechanism of action, targeting expanded CAG repeats in the mutant transcript and inhibiting translation of the mutant allele via steric block. It is currently in phase 1/2a clinical trials, and the company announced positive interim biomarker data in September 2024.

Meanwhile, in the preclinical space, Sangamo/Takeda are developing a mutant-allele selective approach, focusing on blocking transcription of the huntingtin gene using lentiviral vector delivered zinc finger repressor transcription factors (ZFP-TFs) that target the pathogenic CAG repeat. They have shown that their ZFP-TFs repress >99% of disease-causing alleles while preserving expression of normal alleles in patient-derived fibroblasts and neurons. Lentivirally delivered ZFP-TFs lead to functional improvements in mouse models, opening the door to their potential clinical development.

Haystack is aware of at least three other companies developing therapeutics aimed at reducing the toxic effect of HTTexon1, but details of their programs are scarce. China-based HuidaGene Therapeutics is developing a CRISPR-based gene editing product to fix the mutant allele. Galyan Bio was developing GLYN122, a small molecule directly targeting HTTexon1, but the company seems to have ceased operations. Similarly, Vybion has been developing INT41, a functional antibody fragment against HTTexon1, but its current status is also unclear.

It is sobering that over 150 years’ since the first description of Huntington’s disease, which many think of as the archetypal monogenic disease, that we still lack a definitive understanding of its pathogenic mechanism. We don’t know whether the pathology arises from HTT protein, RNA, DNA or some combination of these. And despite the buzz surrounding HTTexon1, most of the data supporting its relevance to human disease still originates from work in mouse models, which recapitulate only certain aspects of the human disorder. That said, raised levels of HTTexon1 are present in patient brain biopsies, with the longer CAG repeats in individuals with juvenile Huntington’s resulting in higher levels of the truncated transcript.

It will be exciting to follow the progress of UniQure’s AMT-130 as our understanding of where in disease progression, and in which patients, this therapy will be most effective. And beyond HTTexon1, other therapeutics targeting alternative disease pathogenic mechanisms are on the horizon. Last month, Skyhawk Therapeutics reported promising phase 1/2 clinical results for it oral small-molecule splice modifier SKY-0515. Elsewhere, broadening understanding of DNA mismatch repair enzymes and the role of somatic repeat instability in the disease have led to investment in a flurry of startup companies focused on this mechanism. That work is now leading to broader excitement that therapies may become available for other difficult-to-treat triplet repeat diseases like Fragile X syndrome, Myotonic dystrophy type 1 and Friedreich ataxia, as demonstrated by the recent deal between Harness Therapeutics and Ono Venture Investment.

Tags: AI, alzheimers, art, artificial-intelligence, beauty, biology, books, cancer, fitness, food, genetics, health, huntingtons-disease, immune-system, immunotherapy, nutrition, painting, poetry, science, technology, wellness, writing