Machine Learning-based Classification of Diffuse Large B-cell Lymphoma Patients by Their Protein Expression Profiles.

Area of Science:

• Proteomics
• Cancer Biology
• Mass Spectrometry

Background:

• Molecular characterization of tumors is crucial for understanding cancer.
• Proteomic analysis of signaling pathways can reveal functional cancer aberrations.
• Accurate tools are needed for robust proteomic characterization of FFPE tissues.

Purpose of the Study:

• To develop a quantitative mass spectrometric pipeline for characterizing formalin-fixed paraffin-embedded (FFPE) diffuse large B-cell lymphoma (DLBCL) tissues.
• To enhance understanding of cancer aberrations at a functional level through proteomic analysis.
• To identify novel protein signatures for DLBCL subtype classification.

Main Methods:

• Combined super-SILAC and label-free quantification (hybrid LFQ) for proteomic analysis.
• Utilized shotgun proteomic analysis on a quadrupole Orbitrap to quantify tumor proteins.
• Applied machine learning (support vector machines) to identify segregating protein candidates.

Main Results:

• Quantified nearly 9,000 tumor proteins in 20 DLBCL patients.
• Successfully segregated DLBCL patients by cell of origin using global protein patterns and a 55-protein signature.
• Identified a panel of four proteins (PALD1, MME, TNFAIP8, TBC1D4) predicted to classify patients with high accuracy.
• Revealed differential expression of core signaling molecules, elucidating DLBCL pathobiology.

Conclusions:

• The developed quantitative proteomic pipeline is effective for characterizing FFPE DLBCL tissues.
• Protein expression patterns can accurately segregate DLBCL subtypes.
• A four-protein panel shows promise for low-error DLBCL patient classification.
• Findings provide insights into the pathobiology of DLBCL subtypes.