VAEG-BIN: Graph Neural Networks for Metagenomic Binning

Data

Each dataset should be in ./data and have the following files:

• node_names.npy
• adj_sparse.npz
• edge_weights.npy
• node_features.npy

If simulated dataset:

• node_to_label.npy
• label_to_node.npy
• labels.npy

If real dataset:

• marker_gene_stats.tsv

The original data can be found at https://zenodo.org/record/6122610

Run command

python train.py <dataset_name>

Note about Results

With the simulated Strong100 dataset, the values calculated during training do not correspond to the final evaluation. For that, use the AMBER tool on the tsv file generated after training with the gold_standard_genome.tsv file. It will generate a file called index.html with the actual AP, AR and F1 values as in the paper. For example:

amber.py  -l "GCN SAGE GAT" -g /host/gold_standard_genome.tsv -o /host/results/ /host/strong100_GCN_vamb0_1l_DIFF-C_results.tsv /host/strong100_SAGE_vamb0_1l_DIFF-C_results.tsv /host/strong100_GAT_vamb0_1l_DIFF-C_results.tsv