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

Table 3 Performance when predicting metabolite (cluster-level) abundance from microbiome (species-level) abundance

Method	Top 10 Corr.Coef	Active links	Stability index
Linear Regression (LR)	0.46	241956	_
+ Sparsity (LASSO)	0.70	6143	0.50
+ Non-neg. weights	0.70	4417*	0.52
CCA	0.47	20449 (E) + 241956 (D)	_
Linear NED	0.60	10010 (E) + 118440 (D)	_
+ Nonlinear activation (NED)	0.74	10010 (E) + 118440 (D)	_
+ Sparsity (Sparse-NED)	0.74	2216 (E) + 4284 (D)	0.79
+ Non-neg. weights (Nonneg-Sparse-NED)	0.74	*1522 (E) + 2568(D)*	0.63

Acronyms: NED neural encoder-decoder; E encoder; D decoder. The stability indices of the fully connected models are naturally 1.0 and not worth including

ISSN: 1471-2164