Sequence features

[DarioMarzella]

For the pilot studies (HLA-A02:01 and HLA-DBR101:01) the following sequence features were used: FullPSSM for the MHC one-hot encoding for the peptide

[DarioMarzella]

03/08/2022 meeting decision: For the whole-allelic experiment ( #41 ):

• Generate one PSSM for every allele sequence, but starting from the same BLAST database. This will provent issues with gaps but will generate virtually identical PSSM s(to check).

Then, the following sequence features will be used: MHC:

• Blosum++/PAM-- encoding
• PSSM_IC Peptide:
• Blosum--/PAM++ encoding OR one-hot encoding
• IC = 0 or very low

Eventually, anchor 1/0 feature can be added to both MHC and pept.

*Note: ++ or -- for blosum and PAM indicate the matrix value. A pam++ is a high pam (e.g. PAM250), and a pam-- is a low pam (PAM50). **Reminder: A high blosum accounts for sequences with high identity. A low pam accounts for sequences with low mutation events between each other (e.g., high identity like a low blosum). On the other hand, a low blosum accounts for sequences with low conservation, so does a high pam.

[gcroci2]

Labels for this issue? @DarioMarzella

[DarioMarzella]

I am not the best at picking labels names, but I hope this is clear enough :)

[heleensev]

23/09/2022 meeting decision:

New PSSM generating proposition:

Data: IPD/IMGT + Uniprot + MGnify (maybe?)

Database generation: HHBlits

MSA generation (per allele): HHBlits

PSSM generation (per allele) PSSMGen

PSSM mapping (per model): PSSMGen