In utils/PDB_processing.py (lines 245-347):
new_pdb_parent = fetch_pdb(pdb_id)
sasa_dic_bound, labeled_seqs = get_sasa_bound(new_pdb_parent)
sasa_dic_unbound, labeled_seqs2 = get_sasa_unbound(new_pdb_parent)
Why are both bound and unbound SASA calculated from the same downloaded PDB file? Also, the MD averaged conformations (frames 15-50) are not used at all. Is this correct?
Paper excerpt:
Ground truth labels: We used a SASA-based method to define interface residues by finding those that are buried in complex form and exposed in isolated form as seen in previous studies. We used a cutoff of 4% change in SASA applied to each conformation to avoid missing transient interactions. Additionally, bound structures come with labels derived from all conformational states, ensuring that interface residues are comprehensively annotated across the dataset.
Surface/core aggregation: Nodes are categorized as either surface or core based on their RSA in the unbound state, using a threshold of 0.1674 to distinguish between buried (core) and exposed (surface) residues.
According to the paper, labels should be derived from SASA changes across each conformation, and RSA should be computed from the unbound state. However, the current code seems to use only a single static PDB file, without utilizing the MD ensemble.
In
utils/PDB_processing.py(lines 245-347):new_pdb_parent = fetch_pdb(pdb_id)
sasa_dic_bound, labeled_seqs = get_sasa_bound(new_pdb_parent)
sasa_dic_unbound, labeled_seqs2 = get_sasa_unbound(new_pdb_parent)
Why are both
boundandunboundSASA calculated from the same downloaded PDB file? Also, the MD averaged conformations (frames 15-50) are not used at all. Is this correct?Paper excerpt:
Ground truth labels: We used a SASA-based method to define interface residues by finding those that are buried in complex form and exposed in isolated form as seen in previous studies. We used a cutoff of 4% change in SASA applied to each conformation to avoid missing transient interactions. Additionally, bound structures come with labels derived from all conformational states, ensuring that interface residues are comprehensively annotated across the dataset.
Surface/core aggregation: Nodes are categorized as either surface or core based on their RSA in the unbound state, using a threshold of 0.1674 to distinguish between buried (core) and exposed (surface) residues.
According to the paper, labels should be derived from SASA changes across each conformation, and RSA should be computed from the unbound state. However, the current code seems to use only a single static PDB file, without utilizing the MD ensemble.