Before launching Tiresias

1) Indexing

The Tiresias framework starts from a provided and/or user-provided. Before starting Tiresias, mapping of the user-provided lists or networks to the provided gene indexing is required. The file our_protein_coding_gene_list_20760genes_all_synonyms_genenames_indexed.tsv includes mapping of HGCN gene names to Ensembl gene IDs to the indexing used in Tiresias. Map the genes from the disease gene list to the protein coding gene list, ending up with the entire protein coding gene list with only the index numbers and the labels corresponding to the disease gene list (0 for non-included genes, 1 for disease genes). Similarly, for the test set of genes which is a set of disease genes of the same disease as the main list but not included in the main one (time-capsule evaluation). All provided networks are already indexed accordingly.

See train_node_labels.tsv as example.

2) Setting up the configuration file

The available computational resources determine the runtime of Tiresias for each selected model. The number of CPUs and/or GPUs, and the GB of RAM available for use need to be stated in the first part of the config.yml file “Configure system resources”. Tiresias will find the input datasets when the config.yml file is also updated with the absolute or relative paths of all input data files in “Set input datasets”. node_attributes and test_node_labels are optional choices for Tiresias, therefore these are filled only if the user uses the gene features provided or their own file (indexed accordingly) or provides a test set. The 7 available models and 11 total model configurations can be appointed for the pipeline. The models are selected in “Enable or disable models”, selecting them by True and disabling them by False.

See config.yml.

3) Setting up the makefile

To launch Tiresias, a command would do the same as selecting the parts of the Tiresias pipeline to run in the Makefile. The entire pipeline includes loading of the data, random walks and embeddings, validation with LOOCV, testing of the selected models, and predicting the ranking of all protein coding genes in the config file. The entire pipeline includes loading of the data, random walks and embeddings, validation with LOOCV, testing of the selected models, and predicting the ranking of all protein coding genes in the config file.

See makefile.