Abstract: TH-OR117
Recognizing Sepsis: A High-Throughput Non-Invasive Assessment Using Machine Learning and Urinary MicroRNAs
Session Information
- Predicting AKI and Clinical Outcomes
October 25, 2018 | Location: 1B, San Diego Convention Center
Abstract Time: 05:42 PM - 05:54 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Kadri, Ferdous, University of Florida, Gainesville, Florida, United States
- Bandyopadhyay, Sabyasachi, University of Florida, Gainesville, Florida, United States
- Adhikari, Lasith, University of Florida, Gainesville, Florida, United States
- Ozrazgat-baslanti, Tezcan, University of Florida, Gainesville, Florida, United States
- Sautina, Laura, University of Florida, Gainesville, Florida, United States
- Lopez, Maria Cecilia, University of Florida, Gainesville, Florida, United States
- Baker, Henry V., University of Florida, Gainesville, Florida, United States
- Segal, Mark S., University of Florida, Gainesville, Florida, United States
- Rashidi, Parisa, University of Florida, Gainesville, Florida, United States
- Bihorac, Azra, University of Florida, Gainesville, Florida, United States
Group or Team Name
- PRISMA-P
Background
Identifying early differences between Systemic Immune Response Syndrome, SIRS, and sepsis remains at the fulcrum of surviving sepsis: without early identification, shock and eventual death ensues. MicroRNAs (miRNAs) are non-coding RNAs that regulate gene expression. Given their selective secretion and extensive role in cellular communication, miRNAs have been utilized in diagnosis, prognosis, and therapeutics in various disciplines of medicine. As the kidneys are one of critical organs affected by sepsis, we surveyed the miRNA population in urine at the time of sepsis. Here we isolate urinary exosomes to identify miRNAs that are associated with sepsis in comparison to patients with vascular disease prior to surgery serving as a control group.
Methods
We collected urine samples from 41 patients within 12 hours of sepsis onset and 29 control patients. Using exosomal isolation protocol (Norgen), we extracted miRNAs from urine supernatant. Subsequently, we utilized Affymetrix® RNA Labeling Kit and GeneChip Arrays to obtain expression values of 2,578 miRNAs in each patient. For machine learning, we randomly divided patients into 70% training set for feature selection of microRNAs and prospectively tested the performance in the validation set (30%) using area under the receiver operating curve (AUC).
Results
94 miRNAs were dysregulated in sepsis (log2FC > 1, p-value < 0.01). 29 miRNAs were highly associated with the septic cohort. Testing the performance of these miRNAs, the validation set performed well with an AUC of 0.94 (95% CI 0.84 - 1.00). Among these, mir-455 is validated to induce hypoxia signaling, whereas mir-3201 upregulates monocyte chemoattractant protein. The mir-548 family is extensively intercalated in the network of TLR signaling, a pathway at the foundation of sepsis.
Conclusion
Uniting machine learning with miRNA arrays allowed us to identify dysregulated miRNAs and peer deeper into the specific pathways activated at the time of sepsis. Actively, we are utilizing this methodology to delineate expression patterns in septic patients as compared to SIRS+ and control groups to further define the fulcrum between SIRS and sepsis.
Funding
- Other NIH Support