Glycan-Targeting Immunotherapy: Where AI Fits Next

Cancer immunotherapy has a blind spot: tumors don’t just hide behind proteins. They also hide behind sugars.

Many cancers rewire the “sugar coating” on their surface—glycans—to signal immune cells to stand down. That’s one reason why checkpoint inhibitors can look spectacular in one patient and disappoint in the next. The immune system may be ready to attack, but it’s being quietly restrained by glyco-immune checkpoints.

A recent research briefing in Nature Biotechnology (published December 16, 2025) spotlights a new modality aimed directly at that restraint: antibody–lectin chimeras (AbLecs), engineered molecules designed to bind and block immunosuppressive glycans. What makes this especially relevant for the AI in Pharmaceuticals & Drug Discovery series is simple: glycans are complex, heterogeneous, and hard to model—exactly the kind of problem where AI can make drug development faster and more predictable.

Why glycans are the immune system’s “quiet off-switch”

Glycans suppress immunity by engaging inhibitory receptors on immune cells. If protein checkpoints like PD-1/PD-L1 are the obvious “brakes,” glycan-mediated signaling is the subtler braking system that many programs under-account for.

The biology pharma teams run into in real tumors

In many tumor microenvironments, immune cells (macrophages, NK cells, some T-cell subsets) express lectin receptors that read glycan patterns. A key family here is the Siglecs (sialic acid–binding immunoglobulin-type lectins). When Siglecs bind sialylated glycans—often elevated on tumor cells—they transmit inhibitory signals that reduce immune activation.

If you’ve worked on immuno-oncology programs, this pattern will sound familiar:

• A tumor shows immune infiltration, but the immune cells are exhausted or nonproductive.
• PD-1 blockade helps… sometimes.
• Resistance mechanisms appear that don’t map neatly to protein checkpoints alone.

Research over the last few years has built the case that the Siglec–sialic acid axis behaves like an immune checkpoint in its own right (including evidence in glioblastoma and humanized models). The practical implication is that protein checkpoint blockade can be necessary but not sufficient.

Why glycans have been hard to drug

Glycans aren’t templated like proteins. They’re built by enzymatic pathways that vary by cell type, metabolic state, and microenvironment. That creates three development headaches:

1. Heterogeneity: the same “target” glycan can differ in linkage and branching.
2. Context dependence: a glycan’s meaning changes depending on what it’s attached to and where it’s displayed.
3. Measurement limits: proteomics has matured; glycomics and glycoproteomics are improving but still less standardized.

This is why many teams have historically focused on protein targets even when glycan biology clearly matters.

AbLecs: antibody precision meets lectin glycan binding

AbLecs are designed to block immunosuppressive glycans by combining an antibody’s targeting specificity with a lectin’s glycan-binding capability. That combination matters because it aims to solve a core challenge: lectins bind glycans well, but they can be too broad; antibodies can be exquisitely specific, but they typically recognize proteins (or a limited set of glycan epitopes).

The research briefing summarizes findings that AbLecs:

• Enhance antitumor immune responses in vitro and in vivo
• Outperform conventional antibody therapies, including approved cancer drugs (in the tested models)
• Act by blocking glyco-immune checkpoints, improving immune control of cancer

A useful mental model is: AbLecs try to localize “glycan blockade” to the tumor context, rather than systemically interfering with glycan biology everywhere.

What’s novel about this modality (from a development standpoint)

Most immunotherapy modalities fall into familiar buckets: monoclonal antibodies, bispecifics, ADCs, cell therapies. AbLecs sit in a different category: a hybrid binder designed around glycan biology.

That changes how teams need to think about:

• Target selection: you’re not just picking a receptor or ligand; you’re picking a glycan motif and its tumor-context presentation.
• Biomarkers: “target positive” can’t rely on a single protein IHC readout; it likely needs glycan-aware assays.
• Safety: glycan motifs can exist in normal tissues, so the therapeutic window depends on tumor enrichment and immune-context effects.

Where AI can accelerate AbLec and glycan-targeting drug discovery

AI is most valuable here because glycan-targeting creates a combinatorial design space that’s too large for intuition and too expensive for brute-force wet lab iteration. In my experience, teams that treat AI as a “model after the fact” leave speed on the table. The better approach is to make AI part of the design loop from day one.

1) AI-guided molecule design for antibody–lectin chimeras

AbLecs introduce more tunable knobs than a standard monoclonal antibody:

• lectin domain choice (and engineered variants)
• binding affinity and avidity balance
• spatial geometry (linker length, orientation)
• Fc region selection and effector function tuning
• developability constraints (stability, aggregation, immunogenicity)

Multi-objective optimization is where AI shines. A practical workflow looks like this:

1. Generate candidate architectures (protein design + domain libraries)
2. Predict binding and specificity against glycan panels
3. Screen for developability risks early (liability motifs, stability, viscosity)
4. Iterate designs using active learning (model suggests next experiments)

Even simple wins matter. If AI helps you cut experimental cycles from, say, 6 rounds to 3 rounds on a lead optimization program, you’re not just saving budget—you’re buying calendar time in a competitive modality.

2) Glycoproteomics and tumor profiling: turning “sugar noise” into features

Glycan data is often treated as messy. AI can turn it into a competitive asset.

• Representation learning can help encode glycan structures (branching, linkages) into model-friendly embeddings.
• Multi-omics integration can connect glycan patterns with transcriptomics (glycosyltransferase expression), proteomics (carrier proteins), and spatial data (tumor vs stroma localization).
• Patient stratification models can identify who is most likely to benefit from glyco-immune checkpoint blockade.

If your commercial team is thinking ahead, this is how you avoid a familiar failure mode: a promising immunotherapy entering trials with an underpowered biomarker strategy.

3) Predicting combination strategies (and avoiding random “combo soup”)

The glyco-immune checkpoint story naturally raises combinations: PD-1/PD-L1, CTLA-4, CD47, macrophage modulators, NK activators.

But combination development can become a “try everything” exercise. AI can impose discipline by:

• modeling signaling networks and immune cell state transitions
• using preclinical and clinical priors to rank combinations
• simulating dose and schedule tradeoffs to reduce toxicity risk

There’s also a strategic point: glycan blockade may shift innate immunity (macrophages/NK cells) as much as adaptive immunity. AI models that include innate pathways tend to make better combo recommendations than T-cell-only frameworks.

4) Clinical trial modeling for a modality with unfamiliar biomarkers

AbLecs may require endpoints and biomarkers beyond the usual PD-L1 IHC and TMB. AI can help trial teams:

• design enrichment strategies using glycan-aware signatures
• model responder probability with real-world-like heterogeneity
• optimize cohort expansion rules based on early biomarker shifts

This is especially relevant heading into 2026 planning cycles: budgets are tight, and leadership wants fewer “science projects” and more programs with credible probability-of-success narratives.

Practical guidance: what to do if you’re evaluating glycan-targeting programs

The fastest way to get value from glycan-targeting immunotherapy is to treat analytics, biomarkers, and design as one system. Splitting them into separate workstreams (discovery now, biomarkers later, AI someday) is how timelines slip.

A diligence checklist for AbLec-style modalities

If you’re on a BD, research, or platform team assessing this space, I’d pressure-test these items early:

1. Target definition: What exact glycan motif is being blocked? How is “positivity” measured?
2. Tumor selectivity: Evidence the glycan signal is enriched in tumor vs critical normal tissues.
3. Mechanism clarity: Which immune cell populations are being disinhibited (macrophages, NK cells, T cells)?
4. Assay robustness: Reproducible glycomics/glycoproteomics pipeline with QC standards.
5. Combination rationale: A mechanistic, testable reason for pairing with PD-1, CD47, or other agents.
6. Developability: Stability, aggregation, and immunogenicity assessment for the chimera format.

“People also ask” (answered plainly)

Are glycans real drug targets or just biomarkers? They’re real targets. The challenge is specificity and context. Modalities like AbLecs exist because teams need a way to intervene in glycan signaling without causing widespread off-tumor effects.

Will glycan targeting replace PD-1 therapies? No. The more realistic path is complementary blockade—glycan checkpoint inhibition addressing resistance mechanisms and improving response depth in selected patients.

Why use AI here instead of standard screening? Because the design space is larger (chimera architecture + glycan specificity + immune effects), and experiments are costlier. AI helps prioritize what to build and test, not just analyze what already happened.

What this means for AI in pharmaceuticals heading into 2026

Glycan-targeting immunotherapy is a strong example of where AI in drug discovery stops being a nice-to-have and becomes the practical path to speed. AbLecs sit at the intersection of complex biology (glyco-immunology), complex molecules (hybrid binders), and complex development (novel biomarkers). That combination tends to overwhelm traditional trial-and-error.

If you’re building an AI strategy for pharma R&D, this is the kind of area worth prioritizing: high unmet need, emerging modality, and a clear reason why data-driven design beats intuition.

A smart next step is to evaluate whether your current stack can handle glycan-aware discovery—data ingestion from glycoproteomics, model architectures that represent glycan structures, and an experimentation loop that closes quickly. If it can’t, you’re likely to watch others set the pace in this corner of immuno-oncology.

Where do you want your pipeline to be by this time next year: running another incremental checkpoint program, or building the capability to develop therapies that tumors can’t “sugar-coat” their way out of?