Fig. 3: AAMB performance on real datasets.
From: Adversarial and variational autoencoders improve metagenomic binning
a Number of NC bins reconstructed from the Almeida and HMP2 datasets for VAMB (blue), AAMB(z) (light green), AAMB(y) (dark green), AAMB(z + y) (light purple), AVAMB (dark purple). b Number of taxa with at least one NC bin genome from VAMB, the de-replicated set of bins generated by AAMB, and the de-replicated set of bins generated by AVAMB from the Almeida dataset. NC bins were classified with GTDB-Tk. Taxa reconstructed by VAMB (blue), taxa reconstructed by AAMB (light purple), and taxa reconstructed by both AAMB and VAMB (dark purple). c Same as B for the HMP2 dataset. d Almeida NC bins quality comparison between VAMB and AAMB. VAMB and AAMB bins were considered equal if belonging to the same sample with 100% identity over at least 75% of the smallest bin. NC: Near complete. V = A: VAMB and AAMB NC bins with exact same score. V > A: VAMB NC bins with a higher score than AAMB NC bins, A > V: AAMB NC bins with a higher score than VAMB NC bins, V unique: VAMB NC bins not reconstructed by any AAMB NC bin at the selected identity settings, A unique: AAMB NC bins not reconstructed by any VAMB NC bin at the selected identity settings. e Same to D for the HMP2 dataset.
