Fig. 3: Results for simulated data with complex motif.
a, Scheme for paired motifs with simulation data. A set of 5 nt was used with two- to six-base insertions and extended to 20 bases. Data containing one motif were also generated. b, Embeddings of split motifs by the VAE. Simulation data generated in a were analyzed using the profile HMM VAE. The resulting embedding plot is shown. Plots generated by the VAE with the multicategorical model decoder are also shown for comparison. c, Representative profile HMM was obtained from the profile HMM VAE. The profile HMM indicated in b is shown. The thickness of the line represents the transition probability. The color of the matching state (filled rectangle) indicates the probability of emitting each nucleotide. A, U, G and C are green, red, yellow and blue, respectively; brown shows the state emitting A, U, G and C equally. The blue rectangles show a high probability of skipping the motif by moving to the deletion state, whereas the red rectangles highlight a high probability of including the motif.
