From Better Spectra to Better Decisions: Five Trends Highlighted at ICORS 2026

The 29th International Conference on Raman Spectroscopy brought the global Raman community to Istanbul from August 23–27, 2026. With almost 300 plenary, keynote, invited, oral, and poster contributions, the ICORS 2026 program offered a broad view of the current state of Raman spectroscopy.

The conference extended from fundamental theory, novel instrumentation, SERS, TERS, and coherent Raman methods to applications in materials science, energy, industry, environmental research, geoscience, planetary exploration, biology, and medicine. This breadth is important: the future of Raman spectroscopy will not be determined by one application or one technological platform.

Opening the meeting, conference chair Mustafa Culha invited participants to “come together to shape the future of Raman spectroscopy.” That future, as the conference demonstrated, is emerging from the interaction of multiple developments: greater analytical reliability and quantification, access to dynamic molecular processes and multimodal imaging, artificial intelligence as a cross-cutting research infrastructure, the expansion of Raman spectroscopy into robust real-world applications, and increasingly integrated translation strategies.

Across these different fields, one overarching transition became visible. Raman spectroscopy is evolving from a technology primarily used to acquire sophisticated spectra into an integrated approach for generating molecular information that can support scientific, industrial, and clinical decisions.

1. Sensitivity is no longer enough: Reliability and quantification

For decades, progress in Raman spectroscopy was frequently described in terms of sensitivity, spatial resolution, or acquisition speed. These parameters remain essential, but ICORS 2026 showed that another performance criterion is becoming equally decisive: reliability.

This shift was especially apparent in surface- and tip-enhanced Raman spectroscopy. Numerous contributions addressed the fabrication and characterization of enhancing structures, quantitative signal evaluation, control of nanoscale probe geometries, and the transfer of measurements between instruments, substrates, and laboratories.

The central question is no longer simply whether an extremely weak molecular signal can be detected. It is whether the same analytical conclusion can be reached on another day, with another instrument, substrate, operator, or sample preparation workflow.

This changes the meaning of technological performance. A record enhancement factor or spatial resolution may demonstrate scientific potential. For many applications, however, reproducible fabrication, calibration, quality control, internal standards, and validated procedures will determine whether that potential can be used in practice because analytical reliability and reproducibility are prerequisites for translating a spectroscopic approach from proof-of-concept studies into robust routine applications.

Quantification is part of the same development. Raman spectroscopy has traditionally been valued for the richness of its molecular fingerprints. Translating these fingerprints into reliable concentrations, material properties, process parameters, or biological states requires stronger links between measurement physics, sample preparation, reference methods, and data analysis.

Work at the Leibniz Institute of Photonic Technology (Leibniz IPHT) in Jena, Germany, on controlled plasmonic structures and stable Raman sensor surfaces illustrates this transition. Its strategic relevance lies not in one substrate architecture alone, but in connecting nanostructure fabrication, spectroscopic characterization, analytical modeling, and application validation.

The reliability race may ultimately prove more important for the wider adoption of Raman spectroscopy than another isolated sensitivity record.

2. From static spectra to dynamics, imaging, and molecular processes

A second major trend was the movement from static molecular characterization toward the observation of spatially and temporally evolving processes.

Coherent Raman techniques, including stimulated Raman scattering and coherent anti-Stokes Raman scattering, continue to increase imaging speed. Time-resolved and ultrafast approaches provide access to excited states, energy transfer, reaction pathways, and transient molecular structures. Raman microscopy and multimodal imaging reveal chemical distributions across surfaces, materials, cells, and tissues.

These developments change the scientific questions that Raman spectroscopy can address. Rather than asking only what molecules or structures are present, researchers can increasingly investigate where they are located, how they interact, and how they change in response to their environment.

This is relevant far beyond biomedicine. Dynamic Raman measurements can follow chemical reactions, phase transitions, catalytic processes, stress and degradation in materials, electrochemical processes, and the behavior of molecular systems under external perturbation. The same principles enable the observation across biological scales. This observation includes metabolism, drug distribution, cellular responses and host–pathogen interactions. In situ and operando measurements are therefore becoming increasingly important. Removing a sample from its functional environment can alter precisely the processes that researchers want to understand. Raman spectroscopy has an advantage here because it can provide chemically specific information with limited sample preparation and, in many cases, without external labels, while Raman tags and bioorthogonal vibrational probes can further extend molecular selectivity when targeted labeling is advantageous.

Multimodal approaches add another dimension. Raman spectroscopy can play a central role in such multimodal strategies because it contributes chemically specific molecular information that is difficult to obtain with many other imaging modalities alone. Fluorescence, electron microscopy, optical coherence tomography, X-ray methods, mass spectrometry, or other analytical techniques can then complement this molecular information with structural, functional, or spatial information.

The future is unlikely to belong to one universally superior modality. It will belong to measurement strategies that combine modalities according to the scientific question and connect their different information levels, with Raman spectroscopy providing a particularly powerful molecular-information layer within such integrated approaches.

3. Artificial intelligence is becoming a cross-cutting research infrastructure

Artificial intelligence was one of the most visible cross-cutting topics at ICORS 2026. In his plenary lecture, Siva Umapathy of the Indian Institute of Science in India placed this development in a broad scientific and societal context. As he summarized in his conference contribution, “AI and deep learning frameworks address these limitations by performing automatic feature extraction, handling nonlinear spectral relationships, and delivering robust classification.”

The important change is that AI is no longer being added only at the end of an experiment to classify spectra. It is beginning to influence the entire Raman workflow: experimental design, measurement optimization, denoising, spectral reconstruction, feature extraction, data fusion, prediction, and uncertainty assessment.

AI is therefore becoming more than an analytical add-on: it is an enabling infrastructure for converting the high-dimensional molecular information generated by Raman spectroscopy into quantitative, reproducible, and decision-relevant information. This becomes particularly important when Raman data are combined with other imaging, spectroscopic, physiological, or clinical data in multimodal approaches.

This development affects virtually every application represented at ICORS. Machine learning can help distinguish mineral phases, monitor industrial processes, interpret complex biological samples, analyze materials, identify microorganisms, or extract weak spectral features from large imaging datasets.

At the same time, the conference highlighted the difference between high accuracy within a curated dataset and genuine analytical robustness. Models intended for practical use must cope with changing instruments, acquisition conditions, sample preparation protocols, environmental conditions, and sample populations.

Shuxia Guo, who leads the AI Laboratory for IR at Leibniz IPHT, and Thomas Bocklitz, head of the institute's Photonic Data Science department, emphasized the enabling role of data analysis in this process: “The power of Raman spectroscopy is largely enhanced by machine learning and chemometrics.” In her ICORS contribution, Guo presented a graph neural network that represents Raman spectra through relationships among spectral bands, thereby embedding spectroscopic prior knowledge into the model architecture. The approach addresses replicate and instrumental variability, supports transfer between devices, and offers a route toward more interpretable classification (https://doi.org/10.1002/cmtd.70119).

This is representative of a broader shift in priorities. The next generation of researchers is asking not only how accurately a model performs, but whether it remains valid when the experimental context changes. Transfer learning, domain adaptation, physics-informed modeling, synthetic data, uncertainty quantification, and FAIR data structures are consequently moving to the center of Raman data science.

Trustworthy AI will require more than larger neural networks. It will require well-characterized samples, meaningful metadata, shared reference datasets, transparent validation strategies, and close connections between data science and the underlying physics and chemistry. Furthermore, the explainability of the models must expand: models should not only classify spectra with high accuracy but also reveal why a decision was made.

The long-term significance of AI may therefore lie less in replacing spectroscopic expertise than in making increasingly complex Raman information reproducible, transferable, and usable across scientific disciplines.

4. Raman spectroscopy is becoming a shared language across disciplines

A striking feature of ICORS 2026 was the growing number of contributions from researchers whose primary expertise lies outside traditional spectroscopy. Engineers, materials scientists, biologists, geoscientists, data scientists, and medical researchers are not simply adopting Raman instruments; they are reshaping the questions the technology is expected to answer.

This broadening suggests that Raman spectroscopy is evolving from a specialist analytical method into a shared molecular-information platform across science and engineering. By involving potential users and application experts earlier in the research process, this disciplinary expansion may also accelerate the path toward industrial and societal applications.

The applications presented at ICORS ranged from advanced materials and energy research to industrial analytics, environmental monitoring, cultural heritage, geology, planetary exploration, and the life sciences. These fields are highly diverse, but they create several common requirements.

Measurements must increasingly work with heterogeneous samples, complex backgrounds, variable environmental conditions, and limited control over sample preparation. Instruments may need to operate in production environments, at remote locations, directly inside reaction systems, or in combination with robotic and automated platforms.

In materials and energy research, many contributions moved beyond structural identification toward structure-function relationships. Raman spectroscopy was used to relate chemical composition, crystallinity, strain, defects, interfaces, and phase transitions to properties such as emission, sensing performance, conductivity, catalytic activity, electrochemical behavior, and degradation. Operando measurements can further connect these relationships with the actual function of materials under working conditions.

Combined with artificial intelligence, such structure-function relationships could gain predictive value. Models that connect spectral signatures with functional properties may reveal non-obvious relationships and, in the longer term, support the development of materials with improved performance for energy, sensing, electronics, and other real-world applications.

In industrial settings, the emphasis shifts toward continuous measurements, automated interpretation, process compatibility, and actionable outputs. Here, the most valuable result may not be a complete spectrum, but a reliable warning that a process is deviating from its desired state.

Environmental, geological, and planetary applications present another set of challenges. Samples may be poorly characterized, spatially heterogeneous, or accessible only through compact and remote instrumentation. Raman spectroscopy must then combine molecular specificity with robust hardware, autonomous data processing, and carefully curated reference libraries.

Biological applications face comparable complexity for different reasons: natural variability, dynamic molecular states, fragile samples, and the need to preserve biological context.

Across all these areas, successful application depends on the complete measurement chain. Sampling, instrumentation, calibration, data analysis, reference information, and user interaction must be considered together.

This expansion into real-world environments may become one of the most important drivers of innovation in Raman spectroscopy. It places less emphasis on idealized performance under optimal conditions and more on the ability to generate trustworthy molecular information where decisions actually have to be made.

5. From application to translation: Molecular Photonics as a strategic frontier

Among these diverse developments, biomedical Raman spectroscopy offers a particularly instructive example of where the field may be heading: from individual measurements toward longitudinal molecular information, and from diagnostic classification toward prevention and adaptive intervention.

Biomedical Raman research has often focused on distinguishing diseased from healthy tissue or identifying a pathogen, cell type, or biomarker. Yet biological and clinical states are not fixed categories. Disease develops, therapies perturb biological systems, microorganisms respond to treatment, and individual molecular baselines differ.

A single spectrum recorded at one time point may therefore be less informative than a sequence of measurements that reveals a trajectory.

Timea Frosch, a Raman spectroscopy researcher at Leibniz IPHT whose work focuses on drug action in infected cells, illustrates this perspective with her research on the stage-dependent effects of antimalarial compounds on hemozoin formation. Resonance Raman microspectroscopy can be used not only to detect a molecular target, but also to investigate how drug–target interactions change within intact infected erythrocytes and across different stages of infection. “What excites me is that Raman spectroscopy allows us to study drug–target interactions inside intact infected cells while preserving the biological context in which the drug acts.”

This type of research points toward longitudinal diagnostics: repeated molecular measurements that follow disease development or therapeutic response within the same biological system or individual. The goal is no longer only to assign a diagnostic label, but to determine the direction and rate of change, identify deviations from an individual baseline, and recognize when an intervention may be required.

This perspective can be framed within the broader concept of Molecular Photonics. Molecular Photonics uses light not simply to generate images or spectra, but to acquire, interpret, and ultimately act on molecular information in biological systems. Raman spectroscopy is a central component because it provides chemically specific information without requiring external labels.

The Personalized Optical Digital Twin

A longer-term perspective emerging from this combination of molecular sensing, multimodal data, and longitudinal analysis is the concept of a Personalized Optical Digital Twin, or PODT. Such a twin would not be based on optical information alone. It would combine molecular spectroscopic and imaging data with physical parameters, physiological signals, laboratory findings, medical history, treatment information, and other patient-specific clinical data.

These heterogeneous data streams would have to be integrated longitudinally and interpreted using reliable, explainable, and uncertainty-aware computational models. The objective is not merely to collect more data, but to transform complementary measurements into a dynamic representation of individual health that can support early detection, therapy monitoring, prevention, and adaptive intervention.

I would summarize this broader perspective as follows: “Photonics provides unique molecular information, but a meaningful digital twin of human health requires more: molecular photonic data must be fused with physical, physiological, and clinical data and interpreted reliably, explainably, and over time.”

Photonics is therefore not the complete digital twin. It provides a distinctive molecular sensing layer that can repeatedly update the twin with information that many conventional measurement technologies cannot access. The usefulness of the resulting model, however, will depend on multimodal data fusion, clinical context, uncertainty quantification, interoperability, and prospective validation.

The term Personalized Optical Digital Twin emphasizes the enabling role of photonics, but the underlying concept is intrinsically multimodal. Its realization will require close cooperation among photonics, medicine, physiology, data science, and systems engineering.

A complete PODT is not yet technically or clinically available. Considerable work remains in sensor integration, standardization, data quality, interoperability, clinical validation, cybersecurity, model drift, and uncertainty quantification. Nevertheless, many of the required scientific components were visible at ICORS: molecular imaging, compact Raman systems, SERS-based bioanalysis, AI-supported interpretation, multimodal data fusion, and measurements of treatment response.

Translation requires dedicated infrastructure

Successfully translating these advances into clinical practice requires more than individual sensors, algorithms, or clinical studies. It requires dedicated translational infrastructures in which photonic technologies, physical and physiological measurements, medical data, computational models, clinical validation, regulatory requirements, and industrial implementation can be brought together. Institutional responses to this challenge can take different forms.

A concrete response to this need is taking shape in Jena, Germany. The Leibniz Center for Photonics in Infection Research (LPI) is an open-access translational infrastructure jointly operated by Leibniz IPHT, the Leibniz Institute for Natural Product Research and Infection Biology, Friedrich Schiller University Jena, and Jena University Hospital.

The LPI organizes its capabilities along a diagnostic and therapeutic development pipeline. It connects technology development, biological and medical testing, clinical validation, and transfer through shared processes and defined handover points. A First-in-Patient Unit planned to enter operation in 2026/27 is intended to provide an environment for evaluating new photonic approaches close to clinical practice.

For Raman spectroscopy, this creates the conditions under which promising methods can progress from controlled laboratory studies to clinically relevant samples, comparative validation, patient-oriented workflows, and ultimately medical application.

The LPI is initially focused on infectious diseases, where the need for faster pathogen identification, antimicrobial resistance testing, immune-response monitoring, and therapy guidance is particularly urgent. At the same time, its translation model has wider significance for Molecular Photonics. Technological development, multimodal data integration, biological understanding, clinical validation, and implementation must be treated as parts of one connected process.

In this context, the LPI provides the type of interdisciplinary and translational environment in which key components relevant to a future PODT – from multimodal sensing and data integration to clinical validation – can be developed and critically tested.

From a spectroscopic toolbox to molecular information systems

The strongest message from ICORS 2026 was one of integration across scales, methods, and applications.

SERS and TERS extend sensitivity and spatial resolution. Coherent and time-resolved methods reveal dynamic processes. Raman imaging provides molecular maps. AI converts complex datasets into interpretable information. Compact and robust systems take measurements into real-world environments. Dedicated infrastructures can then support the transition from technological potential to validated application.

Institutions capable of linking these elements across disciplinary boundaries will play a defining role in the next phase of the field. Leibniz-IPHT provides one example of such an integrated approach, connecting photonic instrumentation, nanostructured enhancement, molecular spectroscopy, imaging, data science, and application-oriented system development.

The LPI and the Center for Biophotonic Technology and Artificial Intelligence (CeBAI) illustrate different institutional responses to the same challenge. The LPI provides an open translational pathway for photonic approaches in infection research, including clinical validation, whereas CeBAI brings together biophotonics and artificial intelligence in an international research environment spanning medicine, forensics, and integrated photonics. Both demonstrate that progress increasingly depends on connecting technology development, data science, users, and implementation.

Leadership in this context is not defined by claiming ownership of every component. It is demonstrated by creating the scientific interfaces, shared infrastructures, and translational pathways that enable different disciplines and partners to move innovations toward application.

The next major advance in Raman spectroscopy may therefore not be a single instrument, substrate, or algorithm. It may be the creation of integrated molecular information systems that combine reliable measurement, multimodal data fusion, contextual understanding, and decision-oriented analysis.

ICORS 2026 showed that many of their scientific foundations are already in place.

About the Author

Jürgen Popp is Scientific Director of the Leibniz Institute of Photonic Technology (Leibniz IPHT) and Professor of Physical Chemistry at Friedrich Schiller University Jena. His research focuses on Raman spectroscopy, biophotonics, and Molecular Photonics, combining advanced photonic technologies with artificial intelligence and multimodal data analysis. His work spans biomedical diagnostics and therapy monitoring, infection research, environmental and food analysis, materials characterization, and the translation of molecular information technologies into robust real-world applications.