Identification of novel biomarkers for Alzheimer's disease: A deep learning omics-based approach to drug pair discovery and exploration of potential therapeutic targets

Comprehensive Summary

This study uses advanced blood protein analysis to discover new biomarkers that could help diagnose Alzheimer's disease and predict which patients will decline cognitively. Researchers analyzed plasma samples from 1,005 participants collected over 12 years at the Massachusetts Alzheimer's Disease Research Center, yielding 1,786 total blood draws. They used liquid chromatography mass spectrometry to measure 4,007 different proteins and 36,259 peptides in these samples, representing the largest unbiased deep proteomics study of dementia conducted to date. Linear statistical analysis identified 138 proteins that were significantly different between Alzheimer's disease patients and healthy controls, with 100 proteins elevated and 38 reduced in disease. Proteins such as MBP, BGLAP, and APoD showed promise as diagnostic biomarkers. Machine learning models using Cox regression analysis identified seven additional proteins (CRISPLD2, CLNS1A, BLVRB, SMYD5, PRPS1, SELENBP1, and OXSR1) whose elevated levels were associated with faster cognitive decline over time. The authors also found that GOLPH3 and VGF were associated with slower cognitive decline, suggesting protective roles. These findings reveal multiple biological pathways involved in Alzheimer's disease including mitochondrial dysfunction, extracellular matrix problems, and inflammation.

Outcomes and Implications

This research is important because current blood biomarkers like phosphorylated tau (pTau) are limited in predicting which patients will decline rapidly and understanding why disease progression varies between individuals. Identifying multiple plasma proteins associated with cognitive decline could enable clinicians to predict future progression and guide treatment decisions in individual patients. The study's largest strength is its large sample size and deep protein coverage, which allowed discovery of proteins at low abundance that would be missed by standard methods. However, significant barriers exist before clinical implementation. The cohort is predominantly white, highly educated, and from specialized research centers, limiting generalizability to diverse patient populations underrepresented in dementia research. Additionally, the study was conducted on stored samples collected over many years with variable timing relative to symptom onset, which may not reflect how these biomarkers perform in clinical practice with standardized collection procedures. Validation of the identified proteins in independent cohorts and clinical trials would be needed, likely requiring 3 to 5 years before any plasma protein panel beyond current standards could be used in routine care.