Journal Article

ProtBAG: eleven organ‐specific proteome‐based biological age using CSF proteomics

Cao, Hui, Wen, Junhao, Ko, Sarah, Bai, Wenjia, Toga, Arthur W., Davatzikos, Christos, Reitz, Christiane, Konofagou, Elisa, Zalesky, Andrew, Saadatinia, Mehrshad, Tian, Ye Ella, Wang, Gao, Edmondson, Donald

2025

Overview

Background Recent research1,2 has generated increasing interest in modeling human aging and disease within a multi‐organ framework. Plasma proteomics3 has emerged as a widely used approach for predicting individual chronological age, resulting in the proteome‐based biological age gap (ProtBAG). Here, we used CSF proteomics from the ADNI study to derive 11 organ‐specific ProtBAGs using 2 machine learning (ML) methods. Method CSF proteomics was generated using the SomaScan 7k platform in ADNI, which included 7,008 protein levels from 736 participants (mean age: 73.3 ± 7.4 years; 57% women). Missing proteomics values were imputed using AutoComplete4. Organ‐enriched proteins for 11 organ systems were determined by at least four‐fold higher mRNA levels in the tissue of interest than in other organ tissues. These organ‐enriched proteins were then fit to a linear support vector regression (SVR) and LASSO regression model. Nested random holdout cross‐validation (50 repetitions) was implemented; mean absolute error (MAE) and Pearson’s r were used to evaluate model performance. Result For proteomics imputation, we chose the imputed results with the copy‐mask amount of 0.3 (2% missing rate in data), which led to the best model performance (r 2=0.54). Overall, LASSO and Linear SVR achieved comparable MAE values with only slight differences between the two models. The brain and hepatic showed the lowest MAE values (Linear SVR: 4.56 and 4.52 for the brain and hepatic ProtBAGs; LASSO 4.55 and 4.52 for the brain and hepatic ProtBAGs) (Figure 1). Across the 11 organ systems, MAE values from our analyses, ranging from 4.5 to 6, were in line with previous literature using brain imaging2. In addition, the brain and hepatic showed the highest Pearson’s r values (Linear SVR: 0.62 and 0.64 for the brain and hepatic ProtBAGs; LASSO 0.63 and 0.65 for the brain and hepatic ProtBAGs) (Figure 2). Other organ systems, such as the endocrine, female reproductive system, and male reproductive system, showed relatively inferior model performance. Conclusion This study leverages CSF proteomics data from ADNI to accurately develop 11 organ‐specific ProtBAGs, enriching the organ aging clock framework established in previous literature using plasma proteomics. Future research will investigate the relationship between these ProtBAGs, cognition, and AD progression.

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: Age / Aging / Biomarkers / Brain / Cognition / Gender differences / Hormones / Literary criticism / Machine learning / Proteins / Proteomics / Reproductive system / Values / Women