Biomarker discovery process at binomial decision point (2BDP): Analytical pipeline to construct biomarker panel

Area of Science:

• Computational biology and bioinformatics
• Machine learning in medicine
• Neuroscience and neurodegenerative diseases

Background:

• Clinical diagnosis, prognosis, and stratification often rely on multiple molecular markers due to the complex nature of clinical incidents.
• Integrating multiple biomarkers is crucial for accurately explaining binary clinical outcomes.

Purpose of the Study:

• To introduce and validate a machine learning pipeline, the Biomarker Discovery Process at Binomial Decision Point (2BDP), for systematic biomarker panel curation.
• To demonstrate the efficacy of 2BDP in identifying biomarker panels for distinct stages of Alzheimer's disease (AD).

Main Methods:

• The 2BDP pipeline integrates feature selection, unsupervised model development, and cross-validation.
• Employed Random Forest and logistic regression to identify minimal biomarker panels differentiating Braak stages I, II, and III in Alzheimer's disease cohorts.
• Evaluated panel efficacy using Area Under the Curve (AUC) from Receiver Operating Characteristic (ROC) plots, calculated via two cross-validation methods.

Main Results:

• Successfully identified three distinct biomarker panels for Braak stages I, II, and III of Alzheimer's disease.
• These panels, comprising 2-10 markers (a mix of novel and known AD signatures), achieved an AUC ≥ 0.8, indicating high diagnostic efficacy.
• The identified markers were weighted and linearly combined to effectively explain the binary outcome variable representing disease stages.

Conclusions:

• The Biomarker Discovery Process at Binomial Decision Point (2BDP) is an effective pipeline for curating optimal biomarker panels with high efficacy.
• 2BDP demonstrates capability in identifying robust biomarker signatures for staging neurodegenerative diseases like Alzheimer's disease.
• Despite limitations such as small sample size and lack of distinct training/test sets, the study highlights 2BDP's potential in biomarker discovery.