Researchers in Germany have built a new computer model that can guess how long small molecules will sit inside a test tube before they pass through it. The tool, developed at Friedrich Schiller University Jena alongside partners from Helmholtz Zentrum München and the Technical University of Munich, could change how scientists identify tiny compounds in complex biological samples.
The problem has lasted for decades. Anyone who analyzes drug candidates, environmental samples or metabolic pathways often starts with a jumbled mixture of small molecules. The new method aims to solve a core question about those tests: how long should each molecule stick around before moving past the detector?
The Problem With Small Molecules
Analytical chemistry depends on separation techniques like chromatography. These methods push a mixture through a column, where each compound moves at its own pace. The pace depends on several factors, including the compound’s shape, weight and chemical bonds. Predicting that pace accurately has been difficult.
Researchers often rely on measured data from similar molecules. That approach works some of the time, but it fails when the sample contains unknown compounds or mixtures that have never been tested before. The result is a slow, laborious process of trial and error.
What the New Method Does
The Jena team’s approach uses machine learning to predict retention times directly from a molecule’s structure. The system takes structural data as input and produces a prediction for how long the molecule will linger during the test. That prediction can then be used to design experiments, set up equipment or interpret results.
The method draws on existing knowledge about how molecular structure relates to behavior in chromatographic tests. By training on large datasets, the model learns patterns that humans might miss.
Why This Matters for Drug Discovery
Drug discovery relies heavily on identifying small molecules that bind to targets. Metabolomics looks at the full set of metabolites in a biological sample. Environmental analysis tracks pollutants and other contaminants. In all three fields, the ability to quickly and accurately identify small molecules is critical.
The new method could speed up the early stages of these processes. Instead of waiting for experimental data, researchers could use the model to estimate retention times ahead of time. That reduces wasted effort and speeds the workflow.
The Collaboration Behind the Work
The project brings together three institutions with distinct strengths:
- Friedrich Schiller University Jena — the lead institution.
- Helmholtz Zentrum München — a research center with a strong focus on biomedical and environmental analysis.
- Technical University of Munich — a major engineering and science university.
The collaboration reflects a broader trend in analytical chemistry toward interdisciplinary work. Machine learning is increasingly being applied to problems that once required lengthy empirical studies.
How the Model Is Trained
Machine learning models need data to learn from. The Jena team’s method likely draws on large datasets of known retention times paired with molecular structures. Over time, the model adjusts its internal parameters to match the data more closely.
The exact architecture of the model is not described in the announcement, so it is not possible to say whether it uses neural networks, decision trees or another approach. What is clear is that the model is designed to generalize beyond the training data, making accurate predictions for molecules it has never seen before.
The Limits of the Approach
No model is perfect. The Jena team’s method will have limits, and the announcement does not describe them. Factors like temperature, solvent composition and column type could all affect retention times in ways the model does not fully capture.
There is also the issue of uncertainty. A prediction is not a guarantee. Scientists will still need to confirm the model’s estimates with actual measurements, especially when stakes are high.
What Comes Next
The team’s announcement describes the method as a step forward, but it does not detail further plans. The method’s value depends on how well it performs across a wide range of molecules and conditions. Early results are promising, but independent testing will be necessary to establish reliability.
Key Facts Box
- Institutions involved: Friedrich Schiller University Jena, Helmholtz Zentrum München, Technical University of Munich
- Field: Analytical chemistry, machine learning, small molecule identification
- Goal: Predict retention times from molecular structure
The method is a practical advance. It removes some of the guesswork from designing chromatographic tests. For researchers working with complex biological samples, that is a real gain.
The method does not replace experimental data. It supplements it. The model provides guidance, but confirmation comes from the test itself.
This is a genuine scientific advance in analytical chemistry. The method addresses a problem that has persisted for decades, and it does so with a technology that is now widely available. The Jena team deserves credit for recognizing the gap and building a solution.
The future of analytical chemistry will likely involve more tools like this one. As computing power grows and datasets expand, machine learning will become ever more useful for predicting chemical behavior. The Jena method is a sign of what is coming.
Source material: “AI method predicts retention times of small molecules more reliably,” Phys.org.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.