Skip to main content

Table 5 This table shows the average “out-of-the-box” AUC for binary classifiers trained on 25 randomly sub-sampled training sets

From: Deep in the Bowel: Highly Interpretable Neural Encoder-Decoder Networks Predict Gut Metabolites from Gut Microbiome

Feature Space

CD vs. HC

UC vs. HC

CD vs. UC

antibiotic

IST

mesalamine

steroids

Microbes only

0.944

0.741

0.784

0.822

0.713

0.582

0.693

Latent space only

0.932

0.759

0.764

0.866

0.731

0.646

0.670

Metabolites only

0.944

0.930

0.781

0.811

0.682

0.956

0.739

Microbes and metabolites

0.950

0.933

0.799

0.838

0.696

0.956

0.750

All layers

0.952

0.925

0.818

0.854

0.717

0.948

0.729

  1. In most cases, the latent space classifier performs at least as well as the microbe classifier. Acronyms: CD Crohn’s disease; UC ulcerative colitis; HC healthy control; IST immunosuppressive therapy