Title:	Harnessing Deep Learning to Unravel Molecular Mechanisms and Functional Pathways of Aging in Vascular Diseases Aging
DNr:	NAISS 2025/22-314
Project Type:	NAISS Small Compute
Principal Investigator:	Anqi Lyu <anqi.lyu@gu.se>
Affiliation:	Göteborgs universitet
Duration:	2025-03-01 – 2026-03-01
Classification:	10203
Keywords:

Abstract

Aging is often perceived as a linear decline; however, it manifests in significant spurts at various critical ages[1]. This project aims to investigate these pivotal transitions in vascular health associated with aging-related diseases. We will utilise proteomic data from multiple cohorts, primarily focusing on Scapix, which includes approximately 5,000 patients and tens of thousands of protein molecules sourced from O-link's high-throughput affinity proteomics. Our analysis will employ two complementary approaches. First, we will apply a novel methodology, designed in-house, integrating vertical clustering to trace the trajectories of specific molecules over time. We intend to utilise deep learning, specifically a variant auto-encoder[2], for molecule profiling and to extract features from the latent space of the model. Our primary focus will be on elucidating these extracted features through further bioinformatics analyses such as UMAP to effectively distinguish significant proteins. Additionally, we will implement a horizontal sliding technique to examine the data across the age spectrum. As a secondary tool, we will employ DE-SWAN[3] to systematically distil key biomarkers by sliding along ages to identify the most significant molecules across various window sizes and p-values. The age with the highest number of significant molecules will be highlighted. This dual analytical approach will uncover the dynamic interactions of proteins, ultimately leading to a comprehensive understanding of their roles in the aging process. The expected outcomes of this project are threefold: first, to identify notable clusters[4] of molecules exhibiting pronounced nonlinear changes during human aging, providing insights into the molecular shifts over time. Second, we will conduct Gene Ontology (GO) and KEGG pathway annotation analyses to elucidate the biological pathways associated with these significant proteins, thereby deepening our understanding of their functional roles[5]. Third, we will screen the entire protein pool to identify crucial biomarkers that are integral to the aging process and the onset of age-related vascular diseases. In addition to these primary outcomes, we aim to explore supplementary findings, including pathways implicated in aging, the specific age changes related to CVD, and the correlation with clinical tests. This research may lead to the development of an aging clock model, with validation through mass spectrometry (MS) to ensure robustness and accuracy. [1] Ahadi, S. et al. Personal aging markers and ageotypes revealed by deep longitudinal profiling. Nat. Med. 26, 83–90 (2020). [2] Zhuang, Jiali, Erin N. Smith, and Dorothee Diogo. "Representation learning based on proteomic profiles uncovers key cell types and biological processes contributing to the plasma proteome." medRxiv (2024): 2024-12. [3] Lehallier, B. et al. Undulating changes in human plasma proteome profiles across the lifespan. Nat. Med. 25, 1843–1850 (2019). [4] Shen, X. Multi-omics microsampling for the profiling of lifestyle- associated changes in health. Nat. Biomed. Eng. 8, 11–29 (2024). [5] Zhuang, Jiali, Erin N. Smith, and Dorothee Diogo. "Representation learning based on proteomic profiles uncovers key cell types and biological processes contributing to the plasma proteome." medRxiv (2024): 2024-12.