Structural data for PARAS: Predictive Algorithm for Resolving A-domain Specificity

This upload contains all structural data for the paper PARAS: high-accuracy machine learning of substrate specificities in nonribosomal peptide synthetases.

alphafold_structures contains 4 folders:

full_structures: contains alphafold2 models for 2623 structures with both the N-terminal and the C-terminal A-subdomains of the native protein

n_terminal_structures: contains alphafold2 models for 3300 structures of the N-terminal A-subdomain of the native protein

n_terminal_structures_with_1amu_cterminal: contains alphafold2 models for 3300 structures of the N-terminal A-subdomain of the native protein, combined with the c-terminal domain of the reference structure 1amu from the PDB database

representative_structures_per_substrate: 34 structures, one for each substrate, used for feature importance visualisations in the paper. Mapping to protein IDs can be found in the file representative_structures.txt

It additionally contains the alphaFold model for the TtpD-A3 A domain.

modeller_structures contains 2 folders:

training_structures: 922 structures used for constructing the structure-based alignment used for structure-based feature extraction

train_test_splits: data splits for crossvalidation to assess the difference between structure-based feature extraction and sequence-based feature extraction