Clinically Applicable Deep Learning Algorithm Using Quantitative Proteomic Data.

Area of Science:

• Biomedical data analysis
• Proteomics
• Machine learning applications

Background:

• Quantitative proteomics generates large datasets valuable for biomedical research.
• The utility of deep learning (DL) for identifying biomarkers in proteomics data is not well-established.
• Pancreatic cancer diagnosis requires robust analytical methods for early detection.

Purpose of the Study:

• To evaluate the classification performance of an optimized deep learning (DL) approach for pancreatic cancer diagnosis using quantitative proteomics data.
• To compare the DL method against conventional machine learning and multivariate techniques.
• To identify significant contributing factors for pancreatic cancer classification through DL parameter optimization.

Main Methods:

• Utilized a large dataset of selected reaction monitoring-mass spectrometry (SRM-MS) data from 1008 pancreatic cancer samples.
• Developed and optimized a deep learning (DL) model for data analysis.
• Compared DL performance against Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), k-Nearest Neighbors (k-NN), and Naïve Bayes (NB).

Main Results:

• The DL method achieved the highest classification performance with an Area Under the Curve (AUC) of 0.9472 on the test dataset.
• DL outperformed all conventional methods evaluated in classifying pancreatic cancer from proteomics data.
• Parameter optimization identified key factors contributing to the DL model's accuracy.

Conclusions:

• Deep learning (DL) demonstrates significant advantages in classifying quantitative proteomics data for pancreatic cancer.
• The optimized DL approach shows potential for enhancing the performance of diagnostic assays in clinical practice.
• This study supports the integration of DL into clinical proteomics for improved disease diagnosis.