Notes 20101118 BIOL 614 Presentations - MHC I and MHC 2 Epitope Prediction

From SnOwy - Ed's Wiki Notebook

Speaker:

Intro

• software used: HLA bind, NETMHCII
• significance -- prediction for MHCI and MHCII
• problems discovered in the software

Project

• recognition of an intracellular pathogen
• proteosome in cytosol breaks apart pathogens into small peptides for MHC presentation
• peptides must be presented by MHC to be recognized
• MHCI for intracellular pathogens
• postively selected
• immuniformatics (immunoinformatics ... -- )
  • MHC database
  • Antigen database
  • B-cell and T-cell receptor database
• epitope detection -- important for vaccine development
• discovery of good peptide candidates
• almost too many servers available to MHC peptide binding prediction ...
• quantitative binding methods chosen -- "quantitative matrices"
• peptide binding parameters -- not trivial
• peptide binding database used -- known peptides that bind MHC -- find a known pattern
  • ... QSAR-ish ...
• machine learning methods: ANN and SVM
• structural based method (not highly recommended for this purpose)
• Hla-bind-Parker et al. 1994 -- epitope peptide prediction
  • 154 different peptides
  • 180 different permutations -- 20 amino acids at 9 positions
  • model overfitting!
  • attempted to look for a mathematical regression that minimizes an error function
• methods are used as a prescreening method -- high scoring items are tested in the lab
• MHCI -- IBS hypothesis -- Independent Binding of Sidechains? -- no correlation between any two or more amino acids
• some exceptions!
  • also: MHCII shows stronger patterns of correlation between sites
  • MHCII binding cleft is also open at both ends, epitope peptide may slide around until energy minimized ...
• NetMHCII
  • the quantitative matrix (position weight matrix) -- did not use binding affinity, used IC₅₀
• flanking residues -- on either side of binding motif of epitope
• MHC allele is parameterized
• antigen database agglomerates heterogeneous data and experimental techniques in the literature
• SMM, Gibbs and Tepitope used to predict epitopes that bind in the MHC cleft
  • the above agree on DRB1*0101
  • they do not agree for DRB1*1302
• need for benchmark / protocol / standardiztion of parameters