AbLecs: Targeting Cancer’s Glycan Checkpoints with AI

Immune checkpoint inhibitors are now used in nearly half of US cancer patients, yet fewer than 20% respond—and among responders, at least 25% eventually develop resistance. Those numbers aren’t just sobering; they’re operationally expensive. They represent failed lines of therapy, delayed patient benefit, and a drug development system that still spends too much time betting on targets that are already crowded.

A paper published this week (December 2025) introduces a new immunotherapy modality that goes after a category of targets many teams still treat as “nice biology, hard drugs”: the sugar coat on tumor cells—specifically, glycans that engage inhibitory lectin receptors like Siglec-7 and Siglec-9. The authors call their approach antibody-lectin chimeras (AbLecs).

From an “AI in Pharmaceuticals & Drug Discovery” perspective, AbLecs land at exactly the right moment. Glycobiology is messy. It’s high-dimensional, context-dependent, and historically under-instrumented. That’s precisely the kind of problem where AI-driven drug discovery can separate signal from noise—by mapping glycan patterns, predicting binding and multivalency, and selecting the right tumor antigen “anchor” for localized checkpoint blockade.

Why glyco-immune checkpoints are the next battleground

Cancer doesn’t only hide behind protein checkpoints like PD-1/PD-L1 or CTLA-4. It also exploits glyco-immune checkpoints—interactions between tumor-associated glycans and immune lectin receptors.

The most practical way to think about it:

• Tumors often become hypersialylated (more terminal sialic acids on surface glycans).
• Those sialylated structures engage inhibitory immune receptors, especially Siglecs, on macrophages, NK cells, and some T-cell subsets.
• When Siglecs are engaged at the immune synapse, they dampen Fc receptor signaling, reducing antibody-dependent killing.

This matters because many approved oncology antibodies—trastuzumab, rituximab, cetuximab, and others—depend on Fc-driven effector functions like:

• ADCP (antibody-dependent cellular phagocytosis)
• ADCC (antibody-dependent cellular cytotoxicity)

If Siglecs are effectively pressing the brakes at the same synapse where Fc receptors are trying to press the gas, you get an antibody that binds its antigen but doesn’t translate binding into strong immune clearance.

Why this has been hard to drug

Most teams that try to “drug glycans” run into three recurring issues:

1. Weak immunogenicity of mammalian glycans → hard to raise specific, high-affinity anti-glycan antibodies.
2. Ligand ambiguity → lectin receptors often bind many glycan ligands; the “right” functional ligand can vary by tumor and context.
3. Affinity limits of decoy receptors → soluble lectin-Fc decoys often bind in the µM to mM range, which is not where most therapeutic antibodies live.

So the field has tended to fall back on either:

• untargeted glycan editing (like systemic sialidase-Fc), or
• systemic receptor blockade (anti-Siglec antibodies)

Both can carry risk because they’re not confined to the tumor microenvironment.

What AbLecs are (and why the architecture is the point)

AbLecs are built on a simple but powerful idea: use a high-affinity antibody arm to concentrate a low-affinity lectin domain where it matters.

Mechanistically, AbLecs couple:

• a tumor antigen-targeting antibody (Fab-mediated binding), with
• a lectin receptor glycan-binding domain (a “decoy receptor” arm)

The antibody arm parks the molecule on the tumor surface at nanomolar effective concentrations, and the lectin arm then blocks the tumor’s inhibitory glycans locally.

The study demonstrates this using:

• trastuzumab × Siglec-7 AbLec (T7)
• trastuzumab × Siglec-9 AbLec (T9)

Even though the Siglec decoy arm is low-affinity on its own, the combined molecule binds tumor cells with low-nanomolar apparent KD, comparable to trastuzumab.

Snippet-worthy: AbLecs convert weak glycan binding into potent checkpoint blockade by forcing proximity and multivalency on the tumor surface.

Why “local blockade at the synapse” beats systemic blockade

One of the more opinionated takeaways: most checkpoint programs still think in terms of “block the receptor everywhere.” For glyco-immune checkpoints, that’s a risky default because Siglecs are broadly expressed across innate immunity.

AbLecs flip the strategy:

• they recruit immune cells through FcγR engagement (like standard antibodies), and
• they simultaneously occupy inhibitory glycans right at the immune synapse

In imaging experiments, AbLecs prevent Siglec-7 accumulation at synapses, which is likely why they outperform combinations like “antibody + Siglec-blocking antibody” in functional assays.

What the data says: stronger ADCP/ADCC and in vivo tumor control

The paper’s results are unusually coherent across assays, which is not always the case for new biologic architectures.

In vitro: better phagocytosis and cytotoxicity

Across primary human effector cells:

• Macrophages: T7 and T9 AbLecs increase tumor cell phagocytosis compared to trastuzumab alone.
• NK cells: T7 AbLec increases ADCC versus trastuzumab.
• Neutrophils (PMNs): T7 and T9 increase antibody-mediated killing.

Two mechanistic controls make the claim more credible:

• Blocking Fcγ receptors reduces activity, confirming Fc dependence.
• Blocking Siglecs (or removing sialic acids with sialidase) erases AbLec’s advantage, confirming Siglec–sialoglycan dependence.

In vivo: reduced metastatic burden in a humanized model

Mouse models often fail Siglec biology because human and murine Siglecs don’t map neatly.

The authors used a more relevant approach: humanized Siglec-7/9 and human FcγR mice (with murine orthologs knocked out). In a lung colonization metastasis model, the T9 AbLec significantly reduced lung metastatic burden compared to trastuzumab.

The practical implication for translational teams: this platform isn’t just an in vitro phenomenon; it has a path to in vivo validation when the model is engineered to reflect the biology.

Where AI fits: why glycans are an AI-shaped problem

If you want a clean, protein-only target landscape, glyco-immune checkpoints will frustrate you. If you want a target landscape where AI is genuinely useful, this is it.

1) AI can map “glyco-signatures” that predict immune escape

Glycosylation isn’t templated like DNA. It’s emergent from enzyme expression, metabolic state, and microenvironment.

AI can help by integrating:

• glycoproteomics / glycomics
• single-cell RNA-seq (glycoenzyme expression)
• spatial omics (where inhibitory glycans sit relative to immune cells)
• clinical outcomes (response vs resistance)

The goal isn’t to perfectly annotate every glycan. The goal is to build predictive signatures: “this tumor is likely to suppress Fc effector function via Siglec-9.”

2) AI can optimize multivalency and geometry, not just affinity

AbLecs work because of cooperativity and effective concentration at the surface. That’s a geometry problem.

The study itself uses computational modeling (multivalent interaction simulation) to explain why AbLecs can outcompete soluble decoy receptors. In industry terms, this points to an AI-assisted optimization loop for:

• linker length and flexibility
• Fc engineering choices (assembly and effector profile)
• antigen selection (surface density and heterogeneity)

Practical stance: For multivalent biologics, AI should be optimizing “synapse physics” as much as binding affinity.

3) AI can guide patient stratification and combination logic

The paper shows synergy between glyco-checkpoint blockade and established checkpoints (for example, combining AbLecs with CD47 blockade improved phagocytosis in donors).

In the clinic, the question becomes: who gets which combination, when?

AI can support decisioning by predicting:

• which inhibitory axis dominates (Siglec-7 vs Siglec-9 vs galectins)
• which effector cells are present (macrophage-heavy vs NK-heavy tumors)
• which tumor antigens are sufficiently expressed to “anchor” the AbLec

That combination logic is exactly where many immuno-oncology programs still rely too heavily on intuition.

What drug discovery teams should do next (actionable)

If you’re in pharma/biotech evaluating AI in drug discovery and wondering how to turn papers like this into a pipeline advantage, here are concrete moves.

1) Treat glycosylation as a first-class biomarker, not a footnote

Build a plan to measure at least one of:

• tumor sialylation levels
• Siglec ligand proxy staining (Siglec-Fc binding assays)
• glycoenzyme expression panels

You don’t need perfect glycan structures to start. You need consistent, scalable proxies.

2) Re-rank tumor antigens for “anchoring efficiency”

For AbLecs, antigen choice is not just about tumor specificity. It’s about surface density and synapse formation.

A practical screen looks like:

• antigen density distribution (including low/heterogeneous expression)
• co-localization with inhibitory glycan signals
• internalization rate (too fast can reduce surface residency)

3) Use AI to design the combination strategy before the first IND meeting

The earlier you formalize combination hypotheses, the less you waste in Phase 1 with “we’ll explore later.”

Start with a prioritized matrix:

1. tumor antigen × glyco-checkpoint axis (Siglec-7, Siglec-9, Siglec-10, galectins)
2. effector mechanism (ADCP-heavy vs ADCC-heavy)
3. combination partner (PD-1/PD-L1, CD47/SIRPα, TIM-3)

Then use modeling to narrow experiments, not expand them.

Where AbLecs could go next

The paper already demonstrates modularity across multiple anchors (HER2, CD20, EGFR) and even dual-checkpoint designs (for example, PD-1 × galectin-9 and PD-L1 × Siglec-7).

The bigger opportunity is strategic: AbLecs are a platform for localized immune checkpoint blockade. If they translate clinically, they could shift the safety-efficacy tradeoff that has haunted systemic checkpoint combinations.

And that ties back to the core theme of this series: AI in drug discovery isn’t just about generating molecules faster. It’s about enabling teams to pursue targets that were previously “too complex to industrialize.” Glyco-immune checkpoints fit that description perfectly.

If your team is exploring AI for target discovery, biologics design, or trial enrichment, AbLecs are a useful test case: can your stack handle biology where the target is a pattern, not a single protein?