Related Experiment Video
Updated: Dec 14, 2025
07:55
Harnessing the Power of MicroRNA Cargoes in Small Extracellular Vesicles Released from Fresh-Frozen Human Brain Sections
Published on: November 8, 2024
711
Using neuronal extracellular vesicles and machine learning to predict cognitive deficits in HIV
Lynn Pulliam1,2, Michael Liston3, Bing Sun3
1Departments of Laboratory Medicine and Medicine, University of California, San Francisco, 4150 Clement St., San Francisco, CA, 94121, USA. Lynn.Pulliam@ucsf.edu.
Journal of Neurovirology
|July 19, 2020
Summary
Predicting HIV-associated neurocognitive disorder (HAND) is possible using plasma neuronal extracellular vesicle (nEV) proteins and clinical data. Machine learning models identified CD4 count, HMGB1, and NFL as key predictors of cognitive impairment.
Area of Science:
- Neuroscience
- Biomarkers
- Machine Learning
Background:
- HIV infection can lead to HIV-associated neurocognitive disorder (HAND), impacting cognitive function.
- Accurate prediction of HAND is crucial for timely intervention and management.
- Plasma neuronal extracellular vesicles (nEVs) offer a potential source of biomarkers for neurological conditions.
Purpose of the Study:
- To predict HAND in HIV-infected individuals using plasma nEV proteins and clinical data.
- To identify key nEV proteins and clinical variables that best predict cognitive impairment.
- To explore the utility of machine learning algorithms in HAND prediction.
Main Methods:
- Collected plasma samples from 60 HIV-infected individuals (22 male, 38 female), with 40 diagnosed with HAND.
- Isolated nEVs using immunoadsorption with a neuron-specific L1CAM antibody.
- Quantified High-mobility group box 1 (HMGB1), neurofilament light (NFL), and phosphorylated tau-181 (p-T181-tau) proteins using ELISA.
- Applied three machine learning algorithms to predict cognitive impairment based on clinical data and nEV proteins.
Main Results:
- Support vector machines demonstrated the best performance among the tested algorithms.
- A model combining selected clinical data with HMGB1 and NFL achieved an area under the curve of 0.82 for predicting cognitive impairment.
- Key predictive features included CD4 count, HMGB1, and NFL.
- Phosphorylated tau-181 (p-T181-tau) was not important for HAND assessment but may help differentiate HAND from Alzheimer's disease (AD).
Conclusions:
- Machine learning, utilizing plasma nEV proteins and clinical data, can effectively predict HAND.
- The combination of specific nEV proteins (HMGB1, NFL) and clinical variables (CD4 count) offers a promising approach for predicting neuronal damage.
- This methodology may aid in differentiating HAND from other neurodegenerative diseases and monitoring therapeutic recovery.
