Three new papers: ReScale4DL, Rxiv-Maker and mAIcrobe

Three papers from the lab came out in September 2026.

ReScale4DL in Nature Communications (doi:10.1038/s41467-026-77930-1). Mariana Ferreira trained common segmentation networks on microscopy images downsampled to 6-50% of their original resolution. Down to a quarter of the original resolution, accuracy either improved (by up to 25% mean IoU) or fell by less than 5%. Lower-resolution images cut storage and inference time, so labs can image faster, store less data and use less computing power.

Rxiv-Maker in Journal of Cell Science (doi:10.1242/jcs.265183). Bruno M. Saraiva led this open-source framework that turns a manuscript written in Markdown into a formatted PDF or Word document, numbering figures and citations and handling cross-references. It can also regenerate figures and recompute reported values from the data, so the text matches the latest analysis. Researchers have already used it for studies of the bacterial cell cycle, filopodia proteomics and zebrafish live imaging.

mAIcrobe in Communications AI & Computing (doi:10.1038/s44488-026-00022-y). António Brito built this open-source framework for bacterial image analysis with the Pinho Lab. It combines deep-learning segmentation (StarDist, CellPose and U-Net) with measurements of cell shape and a neural-network classifier, and works on species from spherical Staphylococcus aureus to rod-shaped Escherichia coli. The paper uses it to detect antibiotic-induced changes in E. coli and cell cycle defects in S. aureus DnaA mutants, and ships Jupyter notebooks for training custom models.

Congratulations to Mariana, Bruno, António and all co-authors.