Fig. 2: Results for simulated data.
a, Scheme of simulation data used for evaluating the decoder models. Ten different motifs with a 10% chance of having nucleotide mutations were randomly extended to 20 bases. b, Embeddings of hypothetical motifs using different decoder models. The simulation data obtained in a were subjected to the VAE with the multicategorical, autoregressive and profile HMM. The resulting latent space is shown, where Dim1 and Dim2 are the first and second axis in the space, respectively. The ELBO is in the right bottom corner with the reconstructed error and Kullback–Leibler divergence. Each motif is plotted with different colors.
