13750

10.30848/PJB2026-6(5)

Low-abundance ion-enhanced MS entropy similarity model based on MLP: Advancing biological mass spectrometry analysis

Jiayao Pan, Ruiyang Li and Yong Li

Traditional similarity methods in small-molecule mass spectrometry are severely hindered by a three-order-of-magnitude signal disparity between high-abundance backbone ions (relative intensity >10%) and low-abundance characteristic ions (relative intensity <1%). To address this limitation, a low-abundance ion–enhanced mass spectrometry entropy (MSE) similarity calculation model based on a multi-layer perceptron (MLP) is proposed. The approach involves four-layer db4 wavelet decomposition, soft-threshold denoising, intensity normalization, and calculation of MSE and statistical features. An MSE-constrained nonlinear function and dual-channel MLP establish spectral peak intensity-dynamic parameter mapping, with backpropagation optimizing parameters to enhance low-abundance ion contribution and suppress high-abundance interference. Validation using the MassBank.us and KUST-MS datasets demonstrate statistically significant performance improvements, with  81.18% (KUST-MS) and 77.27% (MassBank.us) of sample groups achieving t-values greater than 2, and over 50% exhibiting p-values below 0.05.The overall Cohen's d was 0.879, with 88.0% large effect sizes (Cohen's d of 0.8 or higher), and 29.4% extremely large effect (Cohen’s d of 1.5 or higher), confirming dynamic weighting significantly enhances the capability to discriminate structural differences in low-spectral-entropy scenarios. The findings of this research are expected to significantly enhance the accuracy of complex mass spectrometry data analysis, thereby providing efficient technical solutions for applications such as metabolomics biomarker screening, environmental trace pollutant analysis, which is important not only for human health but also for biodiversity conservation and drug impurity identification.

Published

2026 58 6

1352

1363