Admin_Adham
TOPLINE:
A machine learning algorithm used gene expression profiles of patients with gout to predict flares.
METHODOLOGY:
- Researchers analyzed clinical data and RNA expression from the peripheral blood of 174 patients with gout and hyperuricemia that had been collected at week 48 of their participation in the STOP Gout trial, which compared the efficacy of febuxostat and allopurinol.
- The team looked for patterns that predicted gout flares in the ensuing 6 months. They tested six different models against 700-1000 genes, using 10-fold cross-validation to eliminate issues with training the model from random samples.
TAKEAWAY:
- The PyTorch neural network performed best, with an area under the curve of 65%.
- The PyTorch model made 57 correct predictions of flare and 46 correct predictions of no flare. It also made 34 incorrect predictions of no flare and 37 incorrect predictions of flare.
- PyTorch outperformed other models with the best positive predictive value at 65% and the best negative predictive value at 60.8%.
- Some of the models were skewed towards false positives or false negatives, and PyTorch had the most balanced results.
IN PRACTICE:
“This gives us proof that transcriptomic data contains enough evidence to be able to predict gout flares, and a neural network, or a deep learning machine learning model, is one of the best candidates to do so. This [study], as far as we know, is the first to use ‘omics’-enabled machine learning for gout flare prediction, and transcript counts from RNA sequence carries a predictive signal, and this adds value to clinical prediction models,” said study presenter Hussain Aljafer, PhD, graduate assistant in computational bioinformatics research at Michigan State University, East Lansing, Michigan.
SOURCE:
The study was presented at the Gout Hyperuricemia and Crystal Associated Disease Network Annual Research Symposium (G-CAN) 2025.
LIMITATIONS:
The data for the study were drawn from a single clinical trial. The model requires further validation.
DISCLOSURES:
Aljafer reported having no financial disclosures.
