Aggregation of Multimodal ICE-MS Data into Joint Classifier Increases Quality of Brain Cancer Tissue Classification

Abstract

Mass spectrometry fingerprinting combined with multidimensional data analysis has been proposed in surgery to determine if a biopsy sample is a tumor. In the specific case of brain tumors, it is complicated to obtain control samples, leading to model overfitting due to unbalanced sample cohorts. Usually, classifiers are trained using a single measurement regime, most notably single ion polarity, but mass range and spectral resolution could also be varied. It is known that lipid groups differ significantly in their ability to produce positive or negative ions; hence, using only one polarity significantly restricts the chemical space available for sample discrimination purposes. In this work, we have developed an approach employing mass spectrometry data obtained by eight different regimes of measurement simultaneously. Regime-specific classifiers are trained, then a mixture of experts techniques based on voting or mean probability is used to aggregate predictions of all trained classifiers and assign a class to the whole sample. The aggregated classifiers have shown a much better performance than any of the single-regime classifiers and help significantly reduce the effect of an unbalanced dataset without any augmentation.

All ions must serve: The role of various regimes of data acquisition in joint classifier for intraoperative mass spectrometry-based glial tumour identification

“…In combination with the aforementioned limitations, it results in a "small sample size" problem which substantially reduces the accuracy of the classification models. As it was demonstrated earlier, 21 the joint classifier that combines multiple weak learners fitted on the mass spectrometric data of various regimes boosts the classification accuracy significantly but parallel acquisition of multimodal data may slow down the analysis procedure substantially. Ensemble learning methods provide an opportunity to reduce the data set unbalancing impact on the classification model's characteristics.…”

“…The second, validation, data set consisted of 26 biopsy samples with a known tumour cell percentage (ranging from 0% to 100%) from 11 glioblastoma patients. 3,21 It was specifically noted that all samples from the same patient were included only in one of the aforementioned data sets. The data sets were acquired with inline cartridge extraction mass spectrometry approach 10,23 using a hybrid high-and low-resolution LTQ Orbitrap XL ETD mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) as described in a preceding work describing the proposed ensemble learning approach.…”

“…The data sets were acquired with inline cartridge extraction mass spectrometry approach 10,23 using a hybrid high-and low-resolution LTQ Orbitrap XL ETD mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) as described in a preceding work describing the proposed ensemble learning approach. 21 Briefly, the brain tumour tissue sample was placed inside a disposable cartridge arranged in front of the vacuum interface of the mass spectrometer. The high voltage and solvent flow consisted of 3:3:3:1 (vol.)…”

“…Each molecular profile consisted of spectra acquired in eight different regimes of measurementall combinations of (1) positive or negative ion mode; (2) high-(Orbitrap mass analyser, resolution 30,000 FWHM at m/z = 400) or low-resolution (linear ion trap mass analyser); (3) wide (100-2000) or narrow (500-1000) m/z range. 21,24 Data analysis All simulations were performed in a standard machine-learning cross-validation set-up. The spectra were normalised, aligned, and all peaks presented in less than ten scans per diagnosis were deleted.…”

“…The resulting spectra were transformed into a feature matrix with individual scans as rows and peak intensities as columns. 21 For the first data set, the resulting matrix rows were divided into training and testing subsets in a 60/40 ratio, ensuring that scans from the same tissue fragment are always assigned to the same set. For each analysed combination of regimes, 10% of all corresponding rows from the training set matrix were chosen at random and were used for the optimisation of XGboost algorithm metaparameters.…”