The Analytical Scientist
  • Explore

    Explore

    • Latest
    • News & Research
    • Trends & Challenges
    • Keynote Interviews
    • Opinion & Personal Narratives
    • Product Profiles
    • App Notes
    • The Product Book

    Featured Topics

    • Mass Spectrometry
    • Chromatography
    • Spectroscopy

    Issues

    • Latest Issue
    • Archive
  • Topics

    Techniques & Tools

    • Mass Spectrometry
    • Chromatography
    • Spectroscopy
    • Microscopy
    • Sensors
    • Data and AI

    • View All Topics

    Applications & Fields

    • Clinical
    • Environmental
    • Food, Beverage & Agriculture
    • Pharma and Biopharma
    • Omics
    • Forensics
  • People & Profiles

    People & Profiles

    • Power List
    • Voices in the Community
    • Sitting Down With
    • Authors & Contributors
  • Business & Education

    Business & Education

    • Innovation
    • Business & Entrepreneurship
    • Career Pathways
  • Events
    • Live Events
    • Webinars
  • Multimedia
    • Video
    • Content Hubs
Subscribe
Subscribe

False

The Analytical Scientist / Issues / 2026 / November / The Search for Label-Free Glycan Sequencing
Spectroscopy Clinical Technology

The Search for Label-Free Glycan Sequencing

Machine-learned molecular motions connect complex spectra with three-dimensional structure

By Henry Thomas, Jessica Allerton, Jessica Allerton 11/17/2026 6 min read

Share

Glycans are structurally diverse, flexible, and difficult to sequence using existing analytical methods. A new framework called DynaSpec combines Raman spectroscopy, enhanced molecular dynamics, and density functional theory to connect spectral bands with the molecular motions that produce them.

Lead author Varun Dolia explains how DynaSpec moves beyond spectral fingerprinting toward structure-based interpretation – and what will be required to extend the method to complex biological samples.

What initially motivated your team to explore the relationship between glycan structure and Raman spectra?

Our lab has long combined Raman spectroscopy with machine learning for cellular phenotyping, including identifying pathogenic bacteria, predicting antibiotic susceptibility, and detecting pathogens in blood. We have also developed nanophotonic resonators that can enhance Raman signals.

Raman remains attractive because it is label-free, informative, requires minimal sample preparation, and is becoming more affordable and scalable. DynaSpec was our first effort to apply it to individual molecular species rather than whole cells.

We wanted to determine whether Raman could do more than recognize molecular fingerprints. Could it reveal molecular structure directly from a spectrum without labels or amplification?

Glycans presented a clear unmet need because they cannot be sequenced as routinely as DNA. We measured 13 National Institute of Standards and Technology N-glycan standards and distinguished them with greater than 85 percent classification accuracy. However, a classifier can only recognize molecules represented in its training data. It cannot explain why spectra differ or identify an unknown structure.

Our goal was therefore to connect each spectral feature to the three-dimensional molecular motion that produced it.

Why are glycans such a challenging test case?

Glycan diversity occurs at several levels. Mammalian glycans use roughly 10 monosaccharides, compared with the four bases in DNA. The same sugars can connect at different positions, each linkage can have an alpha or beta configuration, and the resulting structures can branch in three dimensions. More than a billion structures are physically possible.

Glycans are also not directly template-encoded and cannot be amplified. Researchers must analyze the molecules themselves, often using limited material.

Our panel included positional isomers such as G1a and G1b. These have identical compositions and masses but differ in which branch carries the galactose. Mass spectrometry cannot distinguish them by mass alone.

Glycans also do not remain in one shape. They occupy ensembles of thermally accessible conformations, and Raman measures the combined contribution of those structures. A calculation based on a single, rigid geometry therefore provides an incomplete representation. Any meaningful structure-spectrum map must account for the full conformational ensemble.

How does DynaSpec work?

DynaSpec asks two questions: What shapes does the molecule adopt, and what spectrum does that collection of shapes produce?

First, we use metadynamics to sample molecular conformations. Conventional molecular dynamics can become trapped in one energy well. Metadynamics fills previously visited wells with what we call “computational sand,” pushing the molecule to explore other regions of its free-energy landscape.

For flexible, branched glycans, the coordinates used to guide this process are difficult to select manually. We therefore use a neural network to identify the slowest and most meaningful molecular motions. This produces a free-energy map of the stable conformations and estimates the population of each one.

We then select representative structures from each conformational basin and calculate their Raman spectra using density functional theory. These spectra are combined according to the estimated population of each conformation and compared with the experimental measurements.

Density functional theory also shows how atoms move during each vibrational mode. This allows us to trace a measured spectral band to particular bonds and torsional motions, or begin with a structural feature and predict where it should appear in the spectrum.

In short, metadynamics identifies the molecular shapes and their probabilities, density functional theory translates them into spectra, and experiment tests the predictions.

What did DynaSpec reveal in the 13 N-glycans?

Across the 13-glycan panel, Raman spectra supported greater than 85 percent average classification accuracy, including discrimination among structures differing by only a singular monosaccharide. A single fucose divided the panel into two spectral families, while spectral shape changed progressively with molecular size within each family.

The simulations also revealed a shared conformational switch. Every glycan showed movement of the alpha-1,6-linked branch between an extended “open” state and a compact “curled” state. For G2, the estimated populations were approximately 90 percent open and 10 percent curled.

Population-weighted spectra predicted from the simulations agreed with the experimental spectra, with cosine similarity of approximately 0.8 across peak positions, intensities, and overall shape.

DynaSpec could also connect individual bands with structural features. For example, adding terminal galactose to G0 to form G2 broadened a particular band. The calculations showed that Raman-active modes associated with the new linkage fell within that region.

For the positional isomers G1a and G1b, the calculations predicted that the most useful information would lie between 1,000 and 1,400 cm⁻¹. Restricting the analysis to this window increased discrimination to 71 percent, compared with 68 percent using the full spectrum, and 63 percent using a control window predicted to have minimal torsional sensitivity. The targeted analysis also used three times fewer features.

Were there any results that surprised you?

The conserved open-to-curled switch was unexpected. We initially anticipated molecule-specific behavior, but the same two-state movement appeared in all 13 nonfucosylated, fucosylated, and sialylated glycans. Its presence across the panel suggests that it is a broader property of this molecular class.

We were also surprised by the strength of the reverse structure-spectrum mapping. Rather than examining an experimental spectrum first, we could identify which internal motions distinguished two structures and predict where that difference should appear.

This produced a counterintuitive analytical result. Difficult classification problems are often addressed by giving a model more data. Here, theory showed that excluding much of the spectrum and focusing on the region dominated by glycosidic torsions improved isomer discrimination.

What were the main challenges your team faced, and how did you overcome them?

The main computational challenge was sampling the glycans’ high-dimensional conformational space. Machine-learned collective variables were essential for guiding metadynamics toward the most informative molecular motions.

Computational cost was another constraint. Each density functional theory geometry optimization and frequency calculation required tens of CPU-hours. We analyzed five conformers from each energy basin for every glycan using Stanford’s high-performance computing cluster.

Calculated harmonic frequencies also differ systematically from experimental peak positions. We addressed this with one affine transformation for each glycan class. It was fitted using overall spectral similarity rather than class labels, preventing discriminative information from entering through calibration.

The remaining experimental challenge is sensitivity. Moving from purified reference standards to dilute, heterogeneous biological samples will require substantially stronger signals. We plan to address this using nanophotonic substrates that can enhance Raman measurements by several orders of magnitude.

How could this approach expand the use of Raman spectroscopy?

The objective is to move Raman from empirical fingerprint recognition toward structural analysis. With an ensemble-resolved calculation, spectral bands can be linked to the motion of particular chemical groups. Researchers can ask not only which reference a spectrum resembles, but what the molecule is doing.

Theory-guided spectral window selection also has practical value. Measuring a smaller, informative region could reduce acquisition time and data dimensionality while improving discrimination. It may also help determine which spectral window an instrument needs before it is configured.

The same principle can support mixture analysis. The glycans in our panel include glycoforms monitored on therapeutic antibodies. Using DynaSpec-derived windows, we quantified fucosylation and galactosylation in binary mixtures with R² values above 0.99 using label-free, drop-cast samples.

We do not see DynaSpec as a replacement for mass spectrometry or nuclear magnetic resonance spectroscopy. Raman could instead provide complementary, label-free structural information through a potentially rapid and scalable measurement.

The framework is not limited to glycans. Glycosidic torsions could be replaced with peptide-backbone dihedral angles or polymer-substituent orientations. The workflow remains the same: sample the conformational ensemble, calculate the spectrum, weight each contribution by population, and assign the vibrational modes.

Jennifer Dionne, corresponding author and professor of materials science and engineering at Stanford University.

Our lab asks a fundamental question: can we read a molecule's structure directly from how it vibrates, without labels, without amplification, without knowing in advance what we're looking for? Glycans are among the most difficult molecules on which to test this idea, which is precisely why we chose them.

We see biology as an enormous, mostly unread information source. DynaSpec is intended to help establish the alphabet and grammar needed to interpret it, one molecule and one vibration at a time.

I remember when Varun first showed me the experimental UMAP with the glycan families clearly separated. That's the moment I knew we were onto something. Varun's workflow is what turned that intuition into an actual method for reading molecular structure straight out of a Raman spectrum.

Glycomics has traditionally relied on powerful techniques, such as mass spectrometry and NMR, but resolving closely related isomers and confromational ensembles can still require specialized workflows.  With DynaSpec, the vibrational spectrum itself provides a complimentary source of structural information rather than serving only as a fingerprint matched against a reference library. This represents a different approach to molecular analysis that could potentially be applied beyond glycans.

What are the next steps for DynaSpec?

Computationally, we want to expand from 13 glycans to libraries containing hundreds of structures. That will require greater automation, GPU-accelerated quantum chemistry, and machine-learned surrogate models trained on the current physics-based calculations.

We have already extended the framework computationally to O-glycans and glycosaminoglycans, where it resolves sulfation-dependent conformations and vibrational windows. We have also tested it experimentally with hyaluronic acid. Validation using defined synthetic sulfated fragments is a priority because sulfation patterns are difficult to distinguish by mass spectrometry.

Experimentally, we aim to analyze lower concentrations and more clinically relevant mixtures. Nanophotonic resonators may provide the signal enhancement required. We also want to explore time-resolved measurements and model environmental factors such as surfaces, binding partners, ionic strength, and molecular crowding.

The longer-term goal is de novo molecular identification: providing only a Raman spectrum and recovering the molecule’s structure. This study shows that structural information is present in the spectrum and can be traced to specific molecular features. Combining that interpretability with improved sensitivity could support label-free structural analysis of previously unknown biomolecules.

Newsletters

Receive the latest analytical science news, personalities, education, and career development – weekly to your inbox.

Newsletter Signup Image

About the Author(s)

Henry Thomas

Deputy Editor of The Analytical Scientist

More Articles by Henry Thomas

Jessica Allerton

Associate Editor, The Analytical Scientist

More Articles by Jessica Allerton

Jessica Allerton

Deputy Editor, The Pathologist

More Articles by Jessica Allerton

False

Advertisement

Recommended

False

Related Content

The Analytical Scientist Innovation Awards 2024: #3
Spectroscopy
The Analytical Scientist Innovation Awards 2024: #3

December 6, 2024

4 min read

Bruker’s multiphoton microscopy module, OptoVolt, ranks third in our Innovation Awards. Here, Jimmy Fong, product development lead, walks us through the major moments during development.

More Bang for Your Buck
Spectroscopy
More Bang for Your Buck

December 4, 2024

1 min read

Researchers develop more stable catalysts for dry reforming of methane – a promising method for carbon capture and utilization (CCU)

The Analytical Scientist Innovation Awards 2024: #1
Spectroscopy
The Analytical Scientist Innovation Awards 2024: #1

December 10, 2024

2 min read

And the technology ranked first in our 2024 Innovation Awards is…

The Analytical Scientist Innovation Awards 2024
Spectroscopy
The Analytical Scientist Innovation Awards 2024

December 11, 2024

10 min read

Meet the products – and the experts – defining analytical innovation in 2024

Affiliations:

Specialties:

Areas of Expertise:

Contributions:

False

The Analytical Scientist
Subscribe

About

  • About Us
  • Work at Conexiant Europe
  • Terms and Conditions
  • Privacy Policy
  • Your Privacy Choices
  • Advertise With Us
  • Contact Us

Copyright © 2026 Texere Publishing Limited (trading as Conexiant), with registered number 08113419 whose registered office is at Booths No. 1, Booths Park, Chelford Road, Knutsford, England, WA16 8GS.