The future of carbohydrate chemistry is shifting from craft-glassware manipulations to data-driven closed-loop systems where machine-learning algorithms select the donor, the promoter and even the stir-rate before the chemist picks up a pipette. The three converging vectors in the field toward 2025 horizon are artificial-intelligence models that are trained on diverse glycosylation data sets and homing in on quantitative yield predictions; robotic fluidic platforms that will perform protecting-group choreography autonomously; and engineered enzymes that have been rationally edited with deep-mutational scanning to accept unnatural substrates and to maintain water-compatible mildness. Collectively, these breakthroughs will eliminate the traditional limitations of microheterogeneity, low throughput and empirical optimisation that have relegated glycans to the sidelines of drug-discovery pipelines.
The glycosciences have long suffered from the fine manual skills necessary to assemble monosaccharide building blocks with the desired regio- and stereochemical result. Automation technologies (robotics), cheminformatics (AI) and enzyme engineering are coming together to lift this limitation, to transform glycan synthesis from an artisanal practice into a predictable, cyclic engineering workflow. Automated robotic platforms carry out solid-phase elongation, on-line mass monitoring and real-time feedback-controlled coupling cycles which self-correct when the conversion falls below a desired threshold. AI enters at two complementary levels. Generative models come up with retrosynthetic trees which balance chemical steps against biocatalytic shortcuts. Discriminative models then assign a rank order to the proposed routes based on cost, sustainability and likelihood of success, which the latter are learned from thousands of previous attempts. Enzyme design offers in silico blueprints a complement with on-demand production of tailor-made biocatalysts; directed-evolution campaigns can be guided by active-site hallucination algorithms that propose mutations most likely to tolerate unnatural acceptors or donor substrates. This interplay is iterative: every robotic run returns analytical data to the AI layer, which refines both synthetic plan and enzyme variant for the next iteration. As a result, the traditional demarcation between "chemical" and "enzymatic" steps is fading into a single design space from which the most effective catalyst is selected (whether a small-molecule organocatalyst, an engineered glycosyltransferase, or a hybrid artificial metalloenzyme) based on algorithm, not ideology.
Glycan synthesis under modern conditions is a combinatorial challenge with regards to protecting-group patterns, anomeric leaving groups and activation modes, and doing this labor by hand is cognitively taxing and financially unviable. Here, AI is leveraged by formulating the retrosynthetic problem in a way that is amenable to graph search. We encode each intermediate as a node in the graph, and assign each edge a cost function that factors in anticipated yield, scalability, environmental impact and availability of monosaccharide building blocks. Graph neural networks, such as transformer architectures that are pretrained on the entire corpus of carbohydrate literature, develop a latent space of glycosidic linkages that the model can learn to disconnect in a manner that respects the often-subtle stereoelectronic bias of the anomeric center. We combine this with reinforcement learning: a reward function is defined that, when the proposed route is successfully scaled and executed on a robotic system, accrues a cumulative signal that incrementally discourages the model from proposing a synthesis that is chemically sound but operationally infeasible. Critically, the models are trained in a setting that also includes enzyme-catalyzed reactions, so that hybrid routes, for example a selective enzymatic sialylation followed by a chemical protecting-group reshuffling, are scored alongside fully chemical pathways. Model uncertainty is quantified at every step: when the (posterior) probability of success drops below a user-defined threshold, the system automatically defers to experimental verification rather than issuing a potentially erroneous prediction. The overall effect is to transform the day-to-day lab workflow: instead of beginning with a fixed synthetic dogma, the chemist interrogates an evolving AI dashboard, is presented with a ranked set of scenarios with associated risk factors, literature precedents and carbon-footprint estimates, and only then do they weigh a single reagent.
The inherent stereochemical lability of glycosidic bonds has long been averse to full automation, but modern reactor designs now allow for unattended synthesis of even branched oligosaccharides. The centerpiece is the so-called "glycan printer" – a valve-free microfluidic system in which picolitre-sized droplets of activator, donor, and acceptor are combined in an inert atmosphere with millisecond accuracy, and the resulting plug of reaction is quenched into a cold diluent to prevent side reactions that would otherwise compromise orthogonality. Inline infrared and Raman optical probes monitor anomeric configuration as it develops, and relay a closed-loop control algorithm that dynamically adjusts coupling times as needed. For larger scale syntheses, flow cartridges containing functionalized resins enable catch-and-release purification between elongation steps, bypassing manual silica-gel filtration. Robotic arms move reactors to and from weighing, capping, and analytical workstations while logging all actions to a blockchain-verified ledger for full provenance and auditability upon regulatory review. Most importantly, the entire system is designed to be agnostic of hardware at the software layer: open-source ontologies for glycosylation steps allow protocols developed on one platform, be it an academic piezoelectric dispenser or an industrial continuous-flow apparatus, to be executed verbatim by another. The practical outcome is democratization: a glycan sequence which would have required months of specialist effort can now be produced overnight, with human glycochemists able to shift their focus from repetitive pipetting to design rationale.
Approaches for protein engineering have been evolved over the years from libraries of error-prone PCR up to sculpting active-site topographies for unnatural substrates using computational approaches. Structural hallucination algorithms can be used to design thousands of de novo scaffolds with idealised, transition-state-geometric oxyanion holes, and then use graph neural networks to predict and rank enzyme variants for expression yield, thermostability, and tolerance to the organic cosolvents that are often needed for glycoside solubilization. Active-site volume can be modulated by loop-insertion/deletion algorithms that shift loop regions by 1 amino acid at a time, enabling the enzyme to bind larger glycans without entropic penalty. Chemoenzymatic orthogonalization is designed into the biocatalyst from the start: engineered sulfatases, for example, can be designed to work at slightly acidic pH that matches the pKa optima for downstream β-elimination reactions, facilitating one-pot, two-step conversion. Directed evolution, as a result, is still needed, but less randomly; machine-learning informed starting points target mutational hotspots that are most likely to improve k_cat without sacrificing K_m. Ultra-high-throughput screening of variants can be achieved using droplet microfluidic technologies with throughputs of over 10,000 variants screened per hour, using fluorogenic glycoside analogues that release a fluorescent reporter upon formation of the correct glycosidic linkage. These datasets are then fed back into the generative model, and the cycle of design, synthesis, and evolution is repeated, with each successive iteration becoming increasingly more efficient. The future will likely see more complete "digital twins" of glycan biosynthetic pathways, with multiple engineered enzymes being co-optimized for pathway flux, intermediate solubility, and cofactor recycling, converting glycan production from an artisanal process to a systems-level engineering challenge.
Glycans are rapidly evolving from their historic role as "passive" PK modifiers to new roles as active determinants of therapeutic fate. Monosaccharide codes that can be programmed to interact with innate immune lectins or sculpted glycoprotein micro-heterogeneity to shield neutralizing epitopes allow for the rational design of half-life, cell-selective uptake, and even intracellular signalling intensity. The current 2025 translational landscape is defined by 3 main arenas: mRNA delivery platforms whose lipid bilayers are post-synthetically re-decorated with glycan ligands to confer organ-selective transfection; tumor-associated glycan signatures mined as liquid-biopsy beacons to monitor immunotherapy response in real-time; and therapeutic proteins whose glycans can be enzymatically rewritten to toggle between immune-silent and immune-activating states without changes to the amino-acid sequence. Each arena leverages the chemical orthogonality of carbohydrates whether in the form of bio-orthogonal click handles, metabolic labels, or enzymatic extension to combine biological specificity with engineering pragmatism.
As lipid nanoparticles are delivered intravenously, they naturally distribute to the liver, which is fine for applications such as correcting metabolic enzyme deficiencies. For engineering immune cells, however, off-the-shelf injections are best administered on demand, in order to program an immune response specifically against cancer. Glycan functionalization provides an elegant, simple solution. Sialylated or mannosylated lipid tails are attached to the particle corona so that it acts as a kind of molecular visa to dock at sialic-acid-binding immunoglobulin-like lectins (Siglecs) or mannose receptors, both of which are densely expressed on the surfaces of dendritic cells, macrophages, and B cells. Uptake is then redirected from the liver sinusoids to the spleen and lymph nodes, with the mRNAs inside being expressed in exactly the right immune cell subsets. By controlling the density of the targeting glycans, however, such delivery vehicles avoid receptor-mediated endosomal uptake that would entrap them, and thus degrade their nucleic acid payloads. Compositional versatility also allows display of both a high-affinity targeting glycan and a PEG layer, though the PEG moiety can be designed to be proteolytically cleavable after cell targeting, yielding a programmable "hide and seek" functionality. In animal models, preclinical toxicology studies show glycan-coated nanoparticles to activate less complement than PEGylated nanoparticles, likely because the sugar moieties are recognized as self. The same strategy could be used to in-vivo generate chimeric antigen receptors (CARs) – with mRNAs encoding the chimeric receptors being delivered to T cells, transiently immunizing them against cancer without the ex-vivo cell culture expansion steps. Expanded glycan libraries, guided by machine learning formulation, could eventually be used to achieve even greater cell-type specificity.
Reprogrammed cells exhibit a rewired biosynthetic architecture, producing glycoforms that are uncommon in healthy cells; tumor-associated carbohydrate antigens (TACAs) make up a 'barcode' that is both generic and unique. Inappropriate sialylation, short O-glycans, and over-expression of fucosylated epitopes all distort the cell-surface landscape, establishing ligands that can be interrogated using lectins, antibodies, or glycomimetic aptamers. As glycosylation is non-template based, the resulting heterogeneity embeds information on the micro-environment (eg. hypoxia, nutrient availability, immune pressure) and so serves as a dynamic output of pathogenic progression rather than a static measure. Glycopeptides are now measured using mass-spectrometric workflows that preserve both peptide identity and glycan attachment; by leveraging data-independent acquisition with deep-learning-based quantification, even small changes in branching or linkage can be discerned, long before there is a change in morphology. In the clinic, panels of glycoproteins secreted by the liver have been built as multi-marker assays capable of stratifying patients at risk for hepatocellular carcinoma who can be more closely surveilled even before radiological abnormalities can be seen. On the immunology side of the field, glycan checkpoints act through inhibitory Siglec binding to suppress the activity of cytotoxic lymphocytes; this can be overcome by blocking these glycan-receptor interactions and can be combined with PD-1 blockade, making it conceivable that glyco-epitope quantification could inform immune therapy responsiveness. Glycan biomarkers are not limited to cancers, either. Flares of autoimmune disease are characterized by IgG Fc hypogalactosylation, which directly correlates to their ability to fix complement; this in turn provides a basis for glyco-engineered biologics that can mute disease-driving antibodies. With analytical sensitivity continuing to fall and structural resolution continuing to improve, it is likely that glycan-based diagnostics will transition from specialized lab testing to direct use in clinical decision making, as tools for monitoring minimal residual disease, immune-related adverse events, and emerging resistance mechanisms.
Fig. 1 The importance of glycans can be conceptualised as an extended model of the central dogma.1,5
Glycosylation of recombinant proteins is no longer considered a process oddity to be accepted as a post-translational modification. Recombinant proteins are now often engineered to take advantage of glyco-attachment sites, with the aim to tune their efficacy, safety and manufacturability. For example, the host glyco-machinery can be biased by media supplements, knock-outs or in-vitro remodeling, to enrich for sialylated N-glycans that protect the polypeptide backbone from renal clearance, thus prolonging the half-life in circulation without increasing the administered dose. The opposite can be true by incorporating glycoforms that lack fucose or are bisected by GlcNAc, since both result in higher affinity for FcγRIIIa, and increased antibody-dependent cellular toxicity towards tumor cells, without affecting normal tissue that generally expresses less of the antigens of interest. Other recent advances include extension of half-life through site-specific attachment of glyco-polymers at exposed cysteines that create synthetic "quasi-glycans" of orders of magnitude larger hydrodynamic volume than natural oligosaccharides and a resulting longer circulation time. Synthetic polymers can be also used to append terminal sugars like mannose-6-phosphate that re-route lysosomal enzymes to the affected organ in storage diseases and restore metabolic balance at micro-dose scale. Glyco-modification can be also used to mask peptide epitopes that are recognized by pre-existing antibodies, reducing immunogenicity for drugs that require lifelong administration, like ERT. In the regulatory space, "glyco-critical" quality attributes have also been added to the list, meaning that sponsors are being required to demonstrate batch-to-batch glycan consistency and linking certain glycoforms to pharmacodynamic signals. Continuous bioprocessing and online glycan monitoring are also maturing, and real-time glyco-adjustment within perfusion reactors is becoming a reality, a pre-requisite for being able to dynamically steer the glycosylation profile of therapeutic glycoproteins, instead of accepting them as a product of a stochastic cellular process.
Fig. 2 Overview of mammalian N-glycosylation.2,5
Additive glycan engineering now provides an orthogonal handle for chemists to tune the pharmacological properties of a lead compound, without perturbing the protein backbone. The strategies can involve in vivo enzymatic reshaping of the glycan, chemoenzymatic elongation in vitro and total synthetic neo-glycosylation, enabling a series of structurally diverse glycoforms to be generated for structural and functional evaluation. This can serve to elucidate the role of individual sugars, linkages and branching, and how these features modulate the interplay of a ligand within its endogenous network. Thus, hit optimization can be considered as a multivalent recognition mechanism, rather than a traditional one-to-one ligand–receptor interaction. Modest improvements in affinity may be attributed to changes in entropic or steric effects (such as an increase in dwell time) rather than through tighter binding. Removal of glycans, or negative glycoengineering, can also be used as a tool to eliminate unproductive species at an early stage of development. The latter is of particular relevance as glycan-critical quality attributes continue to be identified in clinical candidates, driving an increasing demand to incorporate glycan design cycles at the hit-to-lead stage.
Historically in SAR-type programs, the sugar component has been considered an accessory ligand, however there is increasing evidence that even minor changes to sugar identity can have a large impact on the overall pharmacological profile. Systematic substitution of each hydroxyl with fluorine or deuterium, for example, has the effect of tuning hydrogen-bond lifetimes while not changing the steric bulk, thus identifying those hydroxyl groups that are kinetic nucleation points for binding to the receptor vs. those that are merely 'solubility placeholders'. Similarly, the anomeric stereochemistry, usually an undesired synthetic liability, can in fact be a conformational toggle that can change the fate of downstream signaling: β→α inversion can cause a receptor to switch from endosomal recycling to lysosomal trafficking, thus truncating or extending overall residence time. Branching patterns are yet another knob for tuning: incorporation of a bisecting N-acetyl-glucosamine, for example, can constrain conformational flexibility of the glycan as a whole, thus reducing the entropy penalty upon binding, and resulting in an overall increase in enthalpic gain. Saturation-transfer NMR can be used to spatially map those sugar protons experiencing intimate contact with the protein surface, thus identifying which arms can be pruned to reduce synthetic complexity while maintaining activity. By cycling through such minor modifications in a semi-defined glycan context, chemists are able to create activity cliffs that are incomprehensible in terms of peptide sequence alone, effectively turning glycosylation from a liability of uncontrolled heterogeneity into a quantitative design element.
Glyco-optimization has become a rational process to balance the often competing requirements for efficacy, safety and developability of antibodies. The Fc glycans sandwiched between the two HC are allosteric modulators of effector function. De-fucosylation reduces the entropic penalty for binding to FcγRIIIa, increasing cellular cytotoxicity to tumor antigens while sparing normal tissues that express the receptor at lower levels. Hyper-sialylation can give an antibody an anti-inflammatory phenotype, and has been used in the context of autoantibody diseases where the goal is to dampen, rather than elicit, the immune system. Site-specific enzyme transfer can move the glycan cargo away from the Fc backbone to an engineered loop in the Fab domain, decoupling antigen binding and effector function, and allowing each to be independently optimized. Heterologous expression in glyco-engineered yeast or moss can also provide human-compatible glycosylation without the risk of adventitious mammalian viruses, as manufacturing supply chains are diversified globally. Glyco-optimisation is also integrated earlier in the discovery process with libraries screened under glycan-defined conditions in early discovery, so that developability metrics such as viscosity, aggregation, and chemical stability, co-evolve with affinity. The molecules produced are often amenable to longer half-lives, lower infusion reactions and greater batch-to-batch consistency, which in turn can allow lower clinical doses and better patient compliance.
The combinatorial complexity of oligosaccharide synthesis has traditionally precluded automation. A new class of microfluidic and photo-cleavable tag approaches now allow thousands of distinct glycosylations to be synthesized and screened in parallel. Iterative glycosylations are performed in nano-scale flow reactors on resin beads that are tagged with photo-labile linkers. At the end of synthesis, the spatially encoded beads are released into 1536-well plates, and individual glycans are interrogated against lectins, antibodies or whole cells. Because sugar recognition is often multivalent, the platform also allows well-defined glycan density gradients, so that avidity can be decoupled from intrinsic affinity. Mass-spectrometric readouts are used to identify non-covalent complexes in real time, and thus identify hits that may have been overlooked by fluorescence-based assays, which can be skewed by inner-filter effects. Machine-learning approaches are then used to interrogate the resulting multi-dimensional data sets—linkage identity, anomeric ratio, spacer length and biological readout—in order to identify hidden pharmacophores for follow-up synthesis. Critically, the whole workflow is closed-loop: failed or otherwise inconclusive reactions are fed back into the liquid-handler scheduling software, which will then modify donor equivalency, temperature ramps or catalyst identity for the next round, thereby collapsing months of empirical optimization into days. The approach opens up access to complex glycan space and allows academic labs to pursue therapeutic hypotheses that were previously limited to specialised centers with large synthetic teams.
Growing convergence of glycoscience with drug discovery and development pipelines is creating strong demand for providers of synthetic glycans. Diversification of the biopharma pipeline away from a near-exclusive focus on protein therapeutics is driving a reassessment of carbohydrates as programmable drug components, rather than sources of interference in analytical assays. Opening up revenue sources well beyond the sale of reagent-grade oligosaccharides, contract developers are supplying cGMP glycan intermediates for applications including mRNA delivery, neo-glycoconjugate vaccines and glyco-optimized antibodies. In addition, there is a growing trend for licensing synthetic platforms for the end-user manufacture of complex sugars, with modular libraries that users can probe for desired structures. Meanwhile, regulatory guidances are increasingly emphasizing glycan identity as a critical quality attribute in biologics manufacturing, driving large pharma to outsource structurally defined carbohydrates rather than depending on variable carbohydrate extracts. This in turn opens up a broadening market space in which specialized providers can add value across the full development continuum, from discovery hit-finding through to commercial API supply if they can show scalable manufacturing capacity, analytical capabilities and are willing to co-innovate in risk-sharing partnerships.
As synthetic glycan chemistry continues to advance, success in research and development depends on access to reliable, high-precision synthesis capabilities. Our glycan synthesis services empower scientists and biopharma innovators to transform conceptual designs into validated, application-ready glycans.
We specialize in custom glycan synthesis—from simple oligosaccharides to highly branched, multifunctional glycans. Our chemists design each synthesis route based on your specific molecular goals, employing optimized chemical or enzymatic methods to achieve superior structural control and reproducibility.
Leveraging automated glycan assembly (AGA), advanced purification systems, and real-time analytical monitoring, we ensure every product meets the highest purity and consistency standards. Each batch is fully characterized using HPLC, LC-MS, and NMR to verify identity and integrity.
Beyond synthesis, we provide integrated analytical and characterization services, allowing your team to streamline workflows and accelerate time to results. Our platform supports applications in vaccine development, antibody glycoengineering, diagnostic assays, and biomaterials research.
Ready to advance your glycan research? Contact our Glycan Synthesis Team to discuss your project needs or request a customized synthesis plan.
1. What are the major trends in glycan chemistry for 2025?
Key trends include automation, AI-driven synthesis planning, enzyme engineering, and the expansion of glycan therapeutics.
2. How is AI used in glycan synthesis?
AI assists in predicting reaction outcomes, optimizing protecting group strategies, and improving automation efficiency.
3. What role does automation play in glycochemistry?
Automated glycan assembly (AGA) accelerates synthesis, enhances reproducibility, and enables large glycan library production.
4. Which industries are investing in glycan innovation?
Biopharma, vaccine developers, and diagnostic companies are leading the adoption of glycan-based solutions for new therapeutics.
References
