Best Manuscript Award

The manuscript

The authors evaluated MS²Rescore, a machine learning-based post-processing tool developed in the CompOmics group in Ghent, across multiple metaproteomics datasets. MS²Rescore enhances classical rescoring by combining search engine-derived features with predicted MS2 peak intensities from MS²PIP and retention times from DeepLC, addressing a long-standing limitation of the field in which low peptide identification rates arise from the large, diverse sequence databases required for multi-species analyses. Using benchmark data from the CAMPI study, four controlled species-mixtures from the iPRG-2020 study, and three large real-world datasets spanning the human gut, biogas plants, and soil, they demonstrated that combining the Sage search engine with MS²Rescore consistently increases identification rates over Sage alone, with the largest gains in the most complex conditions. Critically, MS²Rescore enables lowering the false discovery rate from the commonly used 1% threshold to a stringent 0.1% with minimal sensitivity loss. In the soil dataset, the most analytically challenging environment, identifications at 0.1% FDR even exceeded those of standalone Sage at 5% FDR.

The manuscript further shows that these gains in peptide identification translate directly into more reliable downstream taxonomic annotation. Recognizing that traditional lowest common ancestor analysis remains sensitive to peptide sharing and rare false positives, the authors combined MS²Rescore with the probabilistic Peptonizer2000 framework, producing robust species-level assignments that accurately reflected sample composition. Together, the work makes a strong case for combining machine learning-based rescoring, a stringent 0.1% FDR, and statistical inference to achieve accurate and interpretable metaproteomics analyses.

The ECR authors

Xuxa Malliet is a predoctoral researcher in the CompOmics group at Ghent University in Belgium. She completed her master’s thesis in the same group, under the supervision of Dr. Tim Van Den Bossche, Dr. Christine Carapito, and Prof. Dr. Lennart Martens, and is now continuing this work in her PhD. Her research focuses on developing a computational framework for targeted mass spectrometry assay design, which involves the development of deep learning models for peptide selection and method optimization. Throughout her PhD, she closely collaborates with LSMBO (Laboratoire de Spectrométrie de Masse BioOrganique) at the University of Strasbourg, France, the group of her co-supervisor Dr. Christine Carapito.

Dr. Tim Van Den Bossche is a postdoctoral researcher in the CompOmics group at Ghent University and VIB-UGent Center for Medical Biotechnology in Belgium. He obtained his PhD in Bioinformatics at Ghent University, where he developed computational methods for quantitative metaproteomics and benchmarked them in the CAMPI-1 study. As a bioinformatician, his work combines tool development with metaproteomics metadata standardization. His main current project focuses on the development of mzPeak, a next-generation mass spectrometry data format. Through these activities, he works at the interface of computational method development, scientific coordination, and field-wide standardization.