Plate Nº 88 · recorded October 10, 2026
Chemistry & MaterialsReported finding
AI tool predicts chromatography retention times on unfamiliar instruments
A Jena-led team has built a machine-learning tool, "2-step," that predicts liquid chromatography retention times for small molecules on unfamiliar instruments, without retraining.
By James Calloway3 min read609 words
In brief
- Published in Nature Methods in 2026; DOI 10.1038/s41592-026-03243-2.
- Led by Prof. Dr. Sebastian Böcker at Friedrich Schiller University Jena, with Helmholtz Zentrum München and TU Munich partners.
- Co-first author Fleming Kretschmer developed the method during his doctorate.
- The "2-step" model first predicts a retention order index, then converts it into a retention time using only a few local reference points.
- Released as a software package and web application called "2-step," intended for integration with liquid chromatography and mass spectrometry workflows.
A team led by Friedrich Schiller University Jena bioinformatician Sebastian Böcker has published a machine-learning tool in Nature Methods that predicts liquid chromatography retention times for small molecules, even on instruments it has never seen before.
What problem does the method solve?
In liquid chromatography, scientists push complex mixtures through a separation column. Each molecule is held back to a different degree and exits the column at a different time. That "retention time" acts as a fingerprint. Analysts compare the measured time against a library of values to identify a compound.
Metabolites — small molecules from metabolism, natural products, toxins, drug breakdown products, and many pharmaceuticals — make up the most chemically diverse class in any biological sample. Blood, cell material, or bacterial cultures can each contain thousands of them. Established sequencing handles DNA and proteins well, but small-molecule identification lags behind.
The hitch is that retention times drift with the experimental setup. "The difficulty lies in the fact that retention times depend heavily on the experimental conditions: on the column used, the solvent, the gradient, the pH, the temperature and even on seemingly minor technical changes," Böcker said.
"Even a slightly longer replacement tube can shift measured times significantly." Previous predictors had to be trained on the same measurement system they would later be tested on. That meant researchers had to measure dozens of standard substances first, an expensive and slow step that is often impractical.
How does the two-step approach work?
The Böcker group, working with colleagues at the Helmholtz Zentrum München and the Technical University of Munich, focused on reversed-phase chromatography, the mode used most often in industry and academic labs.
Their method, called "2-step," proceeds in two stages. A neural network first calculates what the team calls a retention order index for a given molecule. This is not a clock reading; it is a position in the expected elution sequence, ranking molecules from earliest to latest. A second step then converts that index into a concrete retention time, using just a handful of local reference measurements.
Fleming Kretschmer, a co-first author of the paper, did much of the work during his doctorate. "A key finding of our work is that our method outperforms other approaches that, unlike ours, first have to be trained extensively on the target system," Kretschmer said. "Our approach therefore enables precise predictions out of the box, even for new systems."
By splitting the job into order and time, the model can transfer what it learned on one instrument to another. The receiving lab only has to supply a few anchors to turn the ranking into a real clock.
Where could the tool be applied?
The need to identify unknown small molecules runs through drug discovery, natural product chemistry, environmental testing, food safety, and pharmaceutical research. Researchers in any of these fields routinely ask the same question: is the measured retention time consistent with the proposed structure?
Natural product chemists screen bacteria, fungi, and plants for antibiotics and cancer drugs. Environmental and food labs test water and crops for contaminants and residues. Pharmaceutical teams verify candidate molecules. In each case, a transferable prediction tool reduces the dependence on instrument-specific calibration.
The team released the method as a software package and a web application under the name "2-step." According to the researchers, it could slot into the analytical software that already drives liquid chromatography and mass spectrometry setups, allowing laboratories to use it automatically during routine runs.
The paper, "Times are changing but order matters: transferable prediction of small-molecule liquid chromatography retention times," appeared in Nature Methods in 2026 (DOI: 10.1038/s41592-026-03243-2).
via Phys.org Chemistry (Source)
More from James Calloway
Show full bio
Staff writer covering marketplaces and e-commerce at SciBeat.
205 articles
Nearby plates
- AI Searches Millions of Enzymes to Find Pollution Fighters
- Could Dead Planets Fake a Biosignature? Complexity Test Holds Up
- Oxford Team Watches Virus-Like Particles Build Themselves in Real Time
- OpenFISH Maps Genes and Metabolites in One Tissue Slice
- Human Milk Responds Selectively to Maternal Diet, Two Studies Find