Research
Research
We are a team of 26 enthusiastic computational biologists, bioinformaticians, and software developers who build methods and tools across the proteomics workflow. Our work spans peptide property prediction, mass spectrometry data analysis, data management, metadata standardization, and the large-scale reuse of public proteomics data.
Modern mass spectrometry measures proteins at a depth and throughput that keeps growing faster than our ability to interpret the resulting data. CompOmics works on that interpretation problem. We develop open-source software, machine learning models, web resources, and data standards that make proteomics data analysis faster and more reproducible. We keep our tools and data resources openly available so that other groups can build on them directly instead of starting from scratch.
Analyte prediction
Proteomics becomes more powerful when we can predict what a peptide should look like before it is measured. We build machine learning models that translate a peptide sequence into its expected behaviour in a mass spectrometer: fragment ion intensities, retention time, and ion mobility. Tools such as MS²PIP, DeepLC, and IM2Deep make these predictions available for everyday analysis workflows, supporting spectral library generation, targeted method design, and identification rescoring. The underlying aim is to make peptide behaviour something we can calculate and test directly, rather than only observe after the fact.
Identification
Peptide and protein identification is usually treated as a database search problem, but much of the real difficulty is ambiguity. A first-pass search engine proposes candidate matches, yet a lot of the information already present in the data goes unused. We develop post-processing and data-driven rescoring methods that combine search engine scores with predicted peptide behaviour, spectrum evidence, retention time, and ion mobility. Tools such as MS²Rescore/TIMS²Rescore, used together with SearchGUI and PeptideShaker, increase the number of confident identifications in DDA and DIA workflows without loosening false discovery control. The point is to recover signal that is already there, not to relax the statistics.
Metaproteomics
Metaproteomics asks which organisms are present in a microbial community and what they are doing, and, just as important, how confidently any of that can be inferred from peptide evidence. That last part is hard: many peptides are shared across related taxa, and many functions are spread across microbial proteomes that are only partly characterized. We build methods for taxonomic and functional interpretation that account for this ambiguity instead of glossing over it. Tools such as Unipept and Peptonizer2000 turn peptide-level evidence into community profiles that are actually interpretable. The goal is to keep metaproteomics statistically grounded at any scale, from a single microbiome study to analyses covering thousands of species.
FAIR & AI-ready data
Public proteomics data only becomes reusable, by people or by machines, when it is well described. We work on metadata annotation, data standards, and community tools that make proteomics data findable, accessible, interoperable, and reusable, and increasingly, ready to train AI models directly. This is a community effort rather than something we do alone. Through PSI-AI we help shape metadata standards for AI use. ProteomicsML gives us a place to curate machine-learning-ready benchmark datasets and teach others how to apply AI to proteomics, and projects such as mzPeak and rusteomics give us fast, open tooling that makes parsing and reusing mass spectrometry data less painful. Resources such as lesSDRF and annotation workflows around public repositories turn deposited datasets into structured objects that can be searched, compared, reprocessed, and used to train models. The underlying goal is to move proteomics data from merely available to actually usable, for new scientific questions and new AI models, beyond the study that originally produced it.
Reprocessing
Reprocessing pulls our other research lines together. Peptide-property prediction improves how we model analyte behaviour, identification and rescoring software recovers confident evidence from complex spectra, and FAIR/AI-ready annotation makes public datasets comparable at scale. We use these pieces together to revisit existing proteomics data with new computational questions and pull out biological information the original analysis did not capture. Recent examples include large-scale tissue profiling, biomarker discovery, and the systematic recovery of post-translational modifications that were missed the first time, as in Scop3PTM. The aim is to keep public proteomics data scientifically active after deposition, so that data already collected can keep producing new findings, not just the ones from its original study.