Skip to main content
Fig. 5 | BMC Bioinformatics

Fig. 5

From: Standard machine learning approaches outperform deep representation learning on phenotype prediction from transcriptomics data

Fig. 5

Performance by unsupervised model and gene set. The binary task performance of each unique model type is shown, grouped by unsupervised model and gene set. A model type is a combination of unsupervised model, supervised model, gene set, and normalization; for example, the recommended model is one model type. Each model type is a single line on this plot. The performance shown is the average of shifted AUCs across binary tasks, weighted by the number of samples in each task to reduce the effect of fluctuations in tasks with fewer samples. There are four unsupervised model types, VAE (variational autoencoder), SDAE (autoencoder), PCA (principal components analysis), and no-embedding (in which the data is unchanged). The best results come from using all genes without an unsupervised embedding

Back to article page