Early Diagnosis: End-to-End CNN-LSTM Models for Mass Spectrometry Data Classification

Area of Science:

• Biomedical data analysis
• Computational biology
• Cancer research

Background:

• Liquid chromatography-mass spectrometry (LC-MS) is vital for cell profiling and cancer research, offering molecular fingerprints of tissues.
• Challenges in LC-MS data analysis include noise, peak shifts, and high dimensionality, hindering accurate cancer diagnosis.
• Deep learning (DL) models can effectively handle complex data, learning features and classifying simultaneously, ideal for raw LC-MS data.

Purpose of the Study:

• To develop an end-to-end deep learning (DL) methodology for analyzing liquid chromatography-mass spectrometry (LC-MS) data.
• To address challenges such as noise, peak shifts, and high dimensionality in LC-MS data for cancer diagnosis.
• To create a DL model capable of early discrimination between tumoral and normal tissues.

Main Methods:

• A novel deep learning (DL) framework combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network was proposed.
• The CNN component reduces data dimensionality and learns spatial features.
• The LSTM component captures temporal patterns within the data.

Main Results:

• The proposed DL model effectively reduces data dimensionality and learns relevant spatial and temporal features from LC-MS data.
• The model demonstrated superior performance compared to benchmark and state-of-the-art models on the same dataset.
• The methodology successfully achieved early discrimination between tumoral and normal tissues.

Conclusions:

• The developed DL framework offers a promising strategy for improving early cancer detection in diagnostic processes.
• The combined CNN-LSTM approach effectively handles the complexities of LC-MS data for cancer profiling.
• This methodology minimizes the need for extensive data preprocessing and feature selection.