Pushing Beyond the Resolution Limits of the "Brick" Miniature Mass Spectrometer Using a Deep Neural Network

In mass spectrometry, achieving high resolution often involves a trade-off with the signal-to-noise ratio (SNR), particularly in miniature mass spectrometers, where both properties typically fall short compared to commercial instruments. This study introduces a novel super-resolution reconstruction algorithm based on bidirectional long short-term memory (Bi-LSTM) networks. This algorithm effectively overcomes hardware performance limitations and reconciles the trade-off between SNR and resolution. The trained Bi-LSTM can perform super-resolution reconstruction of fixed-length mass spectral segments. When comparing the reconstruction results with theoretical ground truth, the normalized Euclidean distance achieves 98.9%, and the cosine similarity for isotope peak ratios reaches 98.1%. The algorithm demonstrates an impressive reconstruction capability for overlapping peaks. Following the integration of a reprocessing method, the algorithm can be directly applied to reconstruct experimental mass spectra. The robustness of the proposed algorithm is further validated through variations in ion trap scanning speed and peak dynamic range. Comparative analyses with a commercial time-of-flight (TOF) mass spectrometer confirm the validity of this method. Specifically, the cosine similarity at the highest scanning speed (6000 Da/s) reaches 0.88, while the cosine logarithmic values for 1 and 10 ng/mL concentrations attain 0.91 and 0.88, respectively. These results underscore the algorithm's effectiveness for analytical testing tasks in miniature mass spectrometers.