Multiple Compounds Recognition from The Tandem Mass Spectral Data Using Convolutional Neural Network.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Analyzing complex mixtures is crucial for identifying compounds in real-world samples.
• Impurities and noise in mixtures can significantly reduce analytical accuracy.
• Traditional methods often require extensive data preprocessing, increasing computational time.

Purpose of the Study:

• To develop a Convolutional Neural Network (CNN) model for analyzing tandem mass spectrometry (MS/MS) data from chemical mixtures.
• To improve the accuracy and efficiency of compound detection in complex samples.
• To reduce the need for extensive data preprocessing steps.

Main Methods:

• A Convolutional Neural Network (CNN) model was developed to analyze MS/MS spectral data.
• The model utilizes a three-channel architecture input to separate and analyze individual compound data.
• The model was trained and tested using data from the Human Metabolome Database (HMDB).

Main Results:

• The CNN model achieved 98% accuracy in analyzing MS data of mixtures from the HMDB.
• Out of 600 test samples, the CNN model achieved 451 true positives, 142 false positives, and 7 true negatives.
• The CNN model significantly outperformed Support Vector Machine (SVM) with Principal Component Analysis (PCA), Deep Neural Network (DNN), Long Short-Term Memory (LSTM), and XGBoost in true positive detection.

Conclusions:

• The proposed CNN model offers a more accurate and efficient approach to analyzing complex chemical mixtures in MS/MS data.
• The CNN model's ability to extract features and classify multi-label spectral data reduces preprocessing burdens.
• Future work will focus on enhancing the model's universality by incorporating data from diverse instruments and additional offset MS data.