![Faculty Proceedings, Presentations, Speeches and Lectures](../../assets/md5images/1cdeaa92fd8267570cc9f6e70a25fde3.png)
Faculty Proceedings, Presentations, Speeches and Lectures
Gulf War Illness Symptom Analysis Mapping from Psychological Evaluations to Cytokines and Hormone Data Using Machine Learning
Event Name/Location
3rd Annual BD2K-LINCS DCIC Systems Biology & Data Science Symposium
Event Location / Date(s)
University of Miami, Miami, FL
Presentation Date
2-1-2018
Document Type
Poster
Description
A third of the Veterans from the 1990-91 Gulf War returned with multiple unexplained chronic symptoms including headaches, joint pain, fatigue, respiratory and cognitive problems, to comprise what is now known as Gulf War Illness (GWI). GWI currently has no treatment, and is therefore an object of current investigation. GWI research is hindered by the heterogeneity of the illness which suggests the existence of multiple subtypes. Here we use Unsupervised Machine Learning methods, with symptom evaluations and the Davidson Trauma Score (DTS) to analyze GWI patients in comparison with Healthy Control Veterans (HC) to determine specific subgroups that can lead to a better understanding of the illness.
Using Expectation Maximization, results show two well defined groups for GWI data (Log Likelihood = -80.4) with only one group for HC data at comparable level of likelihood (Log Likelihood = -79.0). Feature selection showed a clear difference in the DTS for the two GWI groups: one with high values (GWI-H) the other with low values (GWI-L). GWI-H also showed higher scores for fatigue and sickness impact as well as low values for emotional and physical well-being compared to GWI-L. This was expanded upon by applying Supervised Machine Learning algorithms to the identified subgroups to extract specific biosignatures based on hormone and immune signaling molecules. Results between GWI-H and HC showed accuracy of 100% using a Decision Trees method, while GWI-L and HC showed 97.43% accuracy. The resulting decision trees also highlighted the stress hormone cortisol as a prime discriminating factor.
NSUWorks Citation
Castellanos, R.,
Klimas, N. G.,
Craddock, T. J.
(2018). Gulf War Illness Symptom Analysis Mapping from Psychological Evaluations to Cytokines and Hormone Data Using Machine Learning. 3rd Annual BD2K-LINCS DCIC Systems Biology & Data Science Symposium.
Available at: https://nsuworks.nova.edu/cps_facpresentations/3855
COinS