A Deep Learning-Based Tumor Classifier Directly Using MS Raw Data.

Area of Science:

• Proteomics
• Bioinformatics
• Machine Learning

Background:

• Large-scale proteomic profiling using mass spectrometry (MS) generates valuable data for human tumor research.
• Accurate classification of tumor and non-tumor samples is crucial for biomarker discovery, diagnosis, and disease monitoring.
• Traditional MS data analysis faces limitations, including low identification rates, hindering comprehensive analysis.

Purpose of the Study:

• To develop a novel deep learning-based tumor classifier that directly utilizes MS raw data.
• To overcome the limitations of traditional MS data analysis methods that rely on identification and quantification.
• To improve the accuracy and efficiency of distinguishing between tumor and non-tumor samples.

Main Methods:

• Extraction of potential precursor ions with intensities and retention times from MS raw data.
• Development and training of a deep learning-based classifier using the extracted MS data.
• Comparison of the deep learning classifier's performance against traditional machine learning methods.

Main Results:

• The deep learning classifier accurately distinguishes between tumor and non-tumor samples.
• The proposed method demonstrates strong performance, comparable to or exceeding other machine learning approaches.
• The approach is independent of MS data identification and quantification results, addressing traditional limitations.

Conclusions:

• Deep learning applied directly to MS raw data offers a powerful tool for tumor classification.
• This novel strategy can enhance the discovery of potential biomarkers missed by conventional methods.
• The method holds promise for advancing biological and medical research in oncology.