Combined High-Throughput Proteomics and Random Forest Machine-Learning Approach Differentiates and Classifies...

Area of Science:

• Oncology
• Proteomics
• Bioinformatics

Background:

• Colorectal cancer (CRC) is a complex disease with significant inter- and intra-tumor heterogeneity.
• Variability in the tumor microenvironment (TME) further complicates CRC progression and treatment.
• Understanding intra-tumor heterogeneity is crucial for identifying novel therapeutic targets.

Purpose of the Study:

• To investigate intra-tumor heterogeneity in CRC, focusing on metabolic reprogramming and epithelial-mesenchymal transition (EMT).
• To identify novel protein biomarkers and potential therapeutic targets using explorative shotgun proteomics and machine learning.
• To analyze differences in protein expression between deep and superficial tumor regions and non-tumor tissues.

Main Methods:

• Explorative shotgun proteomics was employed to analyze protein expression in deep tumor, superficial tumor, and non-tumor samples (n=16).
• A Random Forest (RF) machine-learning approach was utilized for classification and identification of key proteins.
• Quantitative proteomic data were analyzed to identify significant protein changes and correlations with clinical parameters.

Main Results:

• 91 proteins, including 23 novel potential CRC hallmarks, showed significant quantitative changes among 2009 analyzed proteins.
• An RF model achieved 98.4% accuracy in classifying the three tissue types using a set of 21 proteins.
• Specific proteins like OGDH-E1, sorting nexin-18, and beta-COP were identified as key classifiers for different tumor regions.
• Metabolic reprogramming and EMT-related proteins (e.g., Galectin-3, fibronectin) showed differential expression between tumor regions.
• CRC regions adopted distinct metabolic strategies, including altered glucose metabolism, lipogenesis, and oxidative stress.

Conclusions:

• Intra-tumor heterogeneity significantly influences metabolic reprogramming and EMT in colorectal cancer.
• Proteomics combined with machine learning can accurately classify CRC tissues and identify novel biomarkers.
• Specific proteins and metabolic pathways are differentially regulated in distinct tumor regions, offering potential therapeutic targets.
• Findings correlate with Dukes stage and budding, supporting the identified proteins as potential therapeutic targets for CRC.