Fig. 8: Examples of model performance on the time series of a few select genera from a dataset from the human gut.
From: Predicting microbial community structure and temporal dynamics by using graph neural network models
Left column: The real relative abundances are indicated by the black points alongside the model predictions from both the training (red), validation (blue), and test (green) datasets. The model training and evaluation was done in windows of 10 samples predicting varying number samples into the future. The models used here have been trained on graph network clusters from a dataset from the human gut39. Right column: observed (real historic) values plotted against the predicted values with two different linear regression lines added: A linear regression between the observed and predicted relative abundance values (blue), and a 1.0-slope line (black) to represent the theoretical perfect prediction. The R21:1 values of each model are indicated in the top left corner, calculated based on the residuals of the linear regression model fitted to the data as well as the 1.0-slope line, respectively. Source data is provided as a Source Data Fig. 8.
