greifflab / manuscript_ab_epitope_interaction Goto Github PK

Hi, I have been trying to reproduce the paratope-epitope interaction results with Bio.PDB.NeighborSearch (biopython) considering that residues interact if they have heavy atoms with a distance less than 5A. While the paratope members generally match those presented in frespairs_segment_notationx_len_merged.csv, I am getting consistently larger epitopes.

Some additional checks I did:

• I have found that the IMGT database usually includes those additional epitope sites I am obtaining (I do not know the exact thresholding criteria here, but it is a first test).
• I computed manually some heavy atom distances using PDB 3D view and they coincide with my results.

I provide two particular cases to illustrate what I am trying to convey, although I have encountered scenarios like these on a constant basis (I use the same numbering convention as that of the Akbar et al. 2021 paper):

PDB 1f58

Just focusing on the CDR-H3, I obtain epitope sites: RIHIR (vs RHR in frespairs_segment_notationx_len_merged.csv). In the .csv file, the fourth ILE (pos. 316) is listed as interacting with the CDR-H1, and the second ILE (pos. 314) does not appear at all as an epitope site. However, both of them are listed in IMGT: https://www.imgt.org/3Dstructure-DB/cgi/details.cgi?pdbcode=1f58&Part=Epitope Additionally, if we compute, for example, the distance of two heavy atoms of ILE (antigen pos. 314) and TYR (pos. 100E of heavy chain), we get 3.39A (and these are not necessarily even the nearest heavy atoms):

PDB 1a14

Focusing on the entire heavy chain, I obtain epitope sites: NISIASSNTDWK (vs NIASNTK in frespairs_segment_notationx_len_merged.csv). If we compute, for example, the distance of two heavy atoms of ASP (antigen pos. 402) and ASN (pos. H54 of heavy chain), we get 4.63A (again, these are not necessarily even the nearest heavy atoms):

I wanted to ask if you are considering additional conditions to compute the paratope and epitope members in frespairs_segment_notationx_len_merged.csv, as maybe I am not being able to get the same results because I am missing something important.

Thank you in advance

Hi there,

I tried to find the raw deep learning models following the instruction

Raw deep learning models and outputfiles are also available at https://archive.sigma2.no 
under DOI 10.11582/2020.00060

though 10.11582/2020.00060 was not listed in the search result.

Is there an alternative way to download the deep learning models? Thank you!