On the feasibility of deep learning applications using raw mass spectrometry data

Abstract

Summary In recent years, SWATH-MS has become the proteomic method of choice for data-independent–acquisition, as it enables high proteome coverage, accuracy and reproducibility. However, data analysis is convoluted and requires prior information and expert curation. Furthermore, as quantification is limited to a small set of peptides, potentially important biological information may be discarded. Here we demonstrate that deep learning can be used to learn discriminative features directly from raw MS data, eliminating hence the need of elaborate data processing pipelines. Using transfer learning to overcome sample sparsity, we exploit a collection of publicly available deep learning models already trained for the task of natural image classification. These models are used to produce feature vectors from each mass spectrometry (MS) raw image, which are later used as input for a classifier trained to distinguish tumor from normal prostate biopsies. Although the deep learning models were originally trained for a completely different classification task and no additional fine-tuning is performed on them, we achieve a highly remarkable classification performance of 0.876 AUC. We investigate different types of image preprocessing and encoding. We also investigate whether the inclusion of the secondary MS2 spectra improves the classification performance. Throughout all tested models, we use standard protein expression vectors as gold standards. Even with our naïve implementation, our results suggest that the application of deep learning and transfer learning techniques might pave the way to the broader usage of raw mass spectrometry data in real-time diagnosis. Availability and implementation The open source code used to generate the results from MS images is available on GitHub: https://ibm.biz/mstransc. The raw MS data underlying this article cannot be shared publicly for the privacy of individuals that participated in the study. Processed data including the MS images, their encodings, classification labels and results can be accessed at the following link: https://ibm.box.com/v/mstc-supplementary. Supplementary information Supplementary data are available at Bioinformatics online.

Recent Developments in Machine Learning for Mass Spectrometry

“…By utilizing a larger data set for initially training on a similar “out of domain” task of classifying photos of animals, the model may perform better once fine-tuned for the new “domain” of classifying photos of plants. Transfer learning has already been utilized for MSI, sample classification, , RT prediction, and spectra refinement . However, the repurposing and retraining of domain/task-specific models for MS data as starting points remain limited, although some groups have developed domain/task specific transfer learning approaches from the ground up .…”

Opportunities and challenges for sample preparation and enrichment in mass spectrometry for single‐cell metabolomics

“…The terms ML and DL are often used interchangeably (and not entirely incorrectly so) but there is an important difference: ML requires the user to define features in the data, whereas DL selects these by itself, therefore making the results more objective and allowing for higher throughput [161]. DL has already found its way to the field of proteomics [40,[162][163][164], is suggested to be a valuable tool in the integration of omics data [38,165] and has shown its merit in MS imaging [166]. Furthermore, the increase in ML algorithm performance eventually levels off for increasing data set size, whereas DL algorithm performance keeps improving, thereby surpassing ML algorithm performance.…”

Deep learning-based proteomics enables accurate classification of bulk and single-cell samples

“…As the most prominent example, convolutional neural networks (CNNs) have been applied to classify proteomic samples by converting their peptide profile into a heatmap-like image, distinguishing conditions based on their spatial peak patterns (4)(5)(6)(7)(8). To this end, CNNs rely on binning data of proximate signals to achieve a uniform grid input, and to employ pre-trained networks (4), which aligns individual samples that suffer from variabilities in the measurement (6). Nevertheless, this requires complex pre-processing, and it is unclear to what extent binning affects the data and may cause loss of information.…”