ProMeta: a meta-learning framework for robust disease diagnosis and prediction from plasma proteomics

Area of Science:

• Biomedical data science
• Proteomics
• Machine learning

Background:

• The plasma proteome provides insights into human health, reflecting interactions between genetics and physiology.
• Deep learning in proteomics is limited by the need for large datasets, hindering rare disease research.
• Patient samples for rare or novel diseases are scarce, posing a challenge for traditional machine learning models.

Purpose of the Study:

• To introduce ProMeta, a meta-learning framework designed for robust disease modeling with limited data.
• To enable accurate disease diagnosis and prediction in scenarios with scarce patient samples.
• To overcome the data-scarcity bottleneck in precision medicine for rare diseases.

Main Methods:

• ProMeta integrates knowledge-guided pathway encoding with bi-level meta-optimization.
• It projects proteomic profiles into biologically interpretable functional tokens.
• The framework learns a global initialization with transferable biological priors from large-scale data for rapid adaptation.

Main Results:

• ProMeta outperformed transfer learning and traditional machine learning baselines in disease diagnosis and prediction.
• In 4-shot scenarios (2 cases, 2 controls), ProMeta achieved an AUROC of ~0.69, a 24.6% relative improvement.
• Mechanistic investigation showed ProMeta disentangles cases from controls, acquiring biological rules and identifying disease biomarkers.

Conclusions:

• ProMeta effectively addresses data scarcity in precision medicine.
• The framework offers a scalable and interpretable solution for characterizing diverse human diseases, especially rare ones.
• ProMeta facilitates the development of diagnostic and predictive models even with minimal patient data.