BENCHMARK DATA: PPDbench

[dlemas]

Supplementary Table 4. The statistical information of the additional benchmark
datasets. To evaluate the performance of binary interaction prediction task, we also
generated the negative samples using the same strategy as described in the Methods
section (i.e., shuffling the non-interacting pairs). All these datasets were essentially
derived from the Protein Data Bank (PDB). \Num" is the abbreviation of \Number".

[dlemas]

Supplementary Table 4.

Dataset	Num of PDBs	Num of Proteins	Num of Peptides	Model
PPDbench	133	111	110	4

PPDbench: https://webs.iiitd.edu.in/raghava/ppdbench/dataset.php

• download the 5 different datasets. count the entries to link with the table information above. We can use this to demo our system.

[dlemas]

Update on ppdbench. the files have been distributed to the team. we will start with the ligand files (133). There is a file called ./ppdbench_metadata.csv with all the file names for ligand pdb files.

Next steps to create the PLIP files for benchmark data includes:

• read the file: ./ppdbench_metadata.csv. this file contains all "ligand" pdb files.
• read the directory with ligand pdb files
• load biopython
• read each pdb file and output the sequence as a column in ./ppdbench_metadata.csv
• output the metadata file.

https://biopython.org/wiki/The_Biopython_Structural_Bioinformatics_FAQ

How do I get the sequence of a structure?
The first thing to do is to extract all polypeptides from the structure (see previous entry). The sequence of each polypeptide can then easily be obtained from the Polypeptide objects. The sequence is represented as a Biopython Seq object.

Example:

seq = polypeptide.get_sequence()
print(seq)
Seq('SNVVE...')

The new metadata will then be used to identify the protein chain in the next step.

[dlemas]

1A1M is an MHC

[evanhadam]

The sequences have been inserted into ppdbench_metadata.csv using a request to an API.