raMSI for Ground-Truth Machine Learning of Mass Spectrometry Imaging Data

The application of machine learning (ML) in mass spectrometry imaging (MSI) becomes one of the most eye-catching fields due to the unparallel sensitivity and specificity of MS for molecular detection and the unmatched efficiency and accuracy of ML for pattern recognition. To get ML started, binning is the most common method used to preprocess the MSI data for the acquisition of millions of m/z bins. However, after deep mining of this approach, we find it suffers strikingly serious ambiguity problem, which introduces a fundamental question: was the machine “learning” the intricate MS data properly? In this report, we provide a resolution adaptive method, raMSI, which can attain ground truth molecular features for large datasets and is compatible with different data formats from mainstream mass analyzers. Build on raMSI, a ML ecosystem is designed including data collection, data preparation, database construction, explorative data analysis, modeling and biological insights acquisition. We envision this platform serves for the purpose of motivating cross-disciplinary research involving chemistry, statistics, and biology.