PROTECT: Protein circadian time prediction using unsupervised learning
Year: 2025
Type: article
Abstract: Circadian rhythms regulate human physiology and their disruption is associated with diseases like Alzheimer's disease (AD). Most proteomic datasets lack time labels and face challenges such as small samples, high dimensionality, and noise, hindering circadian analysis. We introduce PROTECT, an unsupervised deep learning method that predicts circadian sample phases without requiring time labels or known rhythmic proteins. Using greedy layer-wise pre-training and cosine-based fine-tuning, PROTECT achieves high accuracy on time-labeled datasets. Applied to unlabeled human proteomic data from postmortem brain regions and urine, PROTECT uncovers circadian disruptions in AD, identifying proteins with retained, lost, or gained rhythmicity. Proteins which retained rhythmicity show region-specific phase shifts and amplitude changes. Enrichment analysis of proteins with altered rhythmicity offers functional insights. This study systematically compares circadian patterns between AD and control subjects using proteomic data, revealing key insights into AD-related circadian dysregulation.
Source: iScience
Authors: Aram Ansary Ogholbake, Qiang Cheng
Institution: University of Kentucky
Language: English
Subfield: Endocrine and Autonomic Systems
Field: Neuroscience
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $3,240
Awards: G3251540411, EAGER: PBI: Modeling Place-based Innovation by Leveraging AI-Enabled Dynamic Graph Techniques, G5934509782, Gene Profile Analysis of Associations between Alzheimer’s Disease and Circadian Rhythmic Patterns of Human Brain Regions Using Machine Learning, RI: Small: Optimal Transport Generative Adversarial Networks: Theory, Algorithms, and Applications, G7291403024, G9000011571, G9000110792
OpenAlex ID: W4415507357