Skip to main content
Fig. 5 | Genome Medicine

Fig. 5

From: Gut microbial determinants of clinically important improvement in patients with rheumatoid arthritis

Fig. 5

Performance evaluation of neural network-based prediction models in determining minimum clinically important improvement and disease activity score (CDAI). a A neural network model was designed to classify patients into one of two MCII patient groups using baseline gut microbiome, clinical, and demographic input features. In leave-one-out cross-validation, this resulted in b a confusion matrix of model predictions showing an overall classification accuracy of 87.5% and a balanced accuracy of 90.0%. MCII+, patients who showed MCII. MCII−, patients who did not show MCII. c A ranked-order of model input features (total: 448) based upon their scaled (from 0 to 1) importance showing that microbiome data were much more influential contributors to the neural network’s decision-making process than clinical and demographic information. Far left: ranked most important; far right: ranked least important. d Another neural network model was constructed to predict CDAI from the same input variables (excluding CDAI) in leave-one-patient-out cross-validation. e CDAI predictions were made on both samples from the same left-out patient in each cross-validation loop (see the “Methods” section). In the scatter-plot, predictions made across all 32 iterations of cross-validation are shown simultaneously. Overall correlation between observed and predicted scores: Spearman’s ρ = 0.37 (P = 0.003; 95% confidence interval: [0.12, 0.58]). Dashed violet line indicates “y = x,” i.e., an exact match between the observed and predicted values

Back to article page