Unsupervised learning reveals novel disease-associated proteins in high-dimensional human proteomic data

Area of Science:

• Biomedical Data Science
• Proteomics
• Computational Biology

Background:

• Precision medicine generates massive proteomic datasets, posing analytical challenges.
• Supervised learning is common but may miss subtle patterns.
• Unsupervised learning can reveal hidden relationships but struggles with high dimensionality.

Purpose of the Study:

• To develop a novel computational framework for analyzing high-dimensional proteomic data.
• To address the limitations of traditional supervised and unsupervised learning methods in large-scale proteomic analysis.

Main Methods:

• Developed the Dimensionality Reduction with Avoidance of Missing/COmmunity Detection (DIRAM/COD) framework.
• Combined dimensionality reduction techniques with unsupervised learning.
• Applied the framework to the UK Biobank proteomic dataset (2,923 proteins, 52,691 participants).

Main Results:

• Successfully analyzed a large-scale proteomic dataset.
• Confirmed established biomarkers for hypertension (UBE2L6) and leukemia (LRCH4).
• Identified novel protein candidates, including IGF2BP3 for celiac disease.

Conclusions:

• The DIRAM/COD framework effectively analyzes high-dimensional proteomic data.
• This approach facilitates the discovery of both known and novel disease biomarkers.
• Opens new avenues for biomarker and therapeutic target identification in precision medicine.