some points to clean up the dataset

#2
by maom - opened
  • split the site column into uniprot_accession and residue_number. In general we want to have one type of data per column, as it makes it easier to join.

  • Add additional columns to identify the target gene_name, uniprot_entry, protein_name "UAP1", "UAP1_HUMAN", "UDP-N-acetylhexosamine pyrophosphorylase". I think you can do this by querying uniprot on the website and exporting it as a tsv, and joining with the dataset on the uniprot_accession. Note the uniprot_accession is precise links specifically a specific version of the protein and is occasionally updated so it's good to have, but it's quite cryptic. The Entry is good in that it has the species and an identifier, but the gene_name is even more general and sometimes more familiar rather than systematic--good to have booth.

  • Since the structural details have to come from somewhere, include a reference to the structure. This is probably best done through adding a pdb_id column.

  • Include a columns for the ligand in the dataset including ligand_smiles, ligand_name and any other columns you think is relevant.

  • In the documentation include for the ligands KB03 etc, a depiction of the molecule, smiles representation.

  • In the documentation, describe what each of the columns mean

  • In the KB03 dataset (for example), rather than having the activity be in the KB03 column, change it to is_active. The idea is that if we have multiple ligandability datasets, we would want to bind them by row in the long form rather than in the wide form. [, , ].

  • Change the val set to validation as it's clearer.

Sign up or log in to comment