1
0The paper introduces CellTune, a specialized software designed to improve the accuracy of cell classification in complex spatial proteomics data. By utilizing a human-in-the-loop active learning workflow, the platform enables researchers to iteratively refine machine learning models through focused manual annotations. The software integrates multichannel visualization, spatial feature analysis, and intuitive labeling tools to overcome common challenges like signal spillover and imbalanced datasets. To support this technology, the authors developed CellTuneDepot, a massive resource containing millions of high-quality labeled cells and tens of thousands of manual annotations. Benchmarking results indicate that CellTune achieves human-level precision, outperforming existing computational methods while facilitating the discovery of novel cell types. Ultimately, this integrated system streamlines the transformation of multiplexed images into detailed, biologically informative cellular maps.
References:
Bussi Y, Shainshein D, Ovits E, et al. CellTune: An integrative software for accurate cell classification in spatial proteomics[J]. Nature Methods, 2026: 1-14.
