Bioinformatic processes for determination of peptide binding
Inventors
Bremel, Robert D. • Homan, Jane
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10706955-B2
Publication Date
2020-07-07
Expiration Date
Abstract
This invention relates to processes for identifying peptide and polypeptide ligands for a binding partner by using principal component analysis of amino acids to derive vectors describing amino acid subsets corresponding to peptides with known binding affinities and then using this information in a neural network modeling process to derive binding prediction equations. These binding prediction equations are then used in the analysis of subsets of amino acids from a target source to identify peptides or polypeptides ligands in the target source that have affinity for a binding partner.
Core Innovation
The invention identifies peptide and polypeptide ligands for a MHC-I or MHC-II binding partner by obtaining an amino acid sequence for a target polypeptide and using a computer to derive peptide binding prediction equations from experimentally derived data comprising a multiplicity of measurements of amino acid physicochemical properties. Principal Component Analysis produces multiple uncorrelated dimensionless, weighted and ranked proxy descriptors that describe at least 80% of the variance in the physicochemical properties of individual amino acids.
The proxy descriptors are used to describe individual amino acids in peptides with known binding affinities to create vectors for peptides with known binding affinities. Neural network modeling derives binding prediction equations for the MHC-I or MHC-II binding partner by comparing the vectors with the known binding affinities, wherein the principal components are utilized as the input layer of a multilayer perceptron neural network.
The process applies the peptide binding prediction equations to a plurality of peptides of defined length from the target polypeptide to predict the ability of the peptides to bind to the binding partner. The neural network determines the permuted average binding of a peptide to at least 20 MHC-I or MHC-II binding regions, displays the binding affinity of the peptides, and selects one or more peptides having a binding affinity of about greater than 10^6 M^-1 to the binding partner.
The selected peptides are cloned into an expression vector for recombinantly synthesizing a biomolecule selected from the group consisting of a protein and nucleic acid encoding the one or more peptides.
Claims Coverage
One independent claim is provided. It combines derivation of MHC-I or MHC-II peptide binding prediction equations from experimentally derived amino acid physicochemical properties using Principal Component Analysis and a multilayer perceptron neural network, prediction of permuted average binding across at least 20 MHC binding regions, and selection of peptides above a specified binding affinity threshold for cloning and recombinant synthesis.
Principal component proxy descriptors from physicochemical properties
Deriving by Principal Component Analysis multiple uncorrelated dimensionless, weighted and ranked proxy descriptors to describe at least 80% of the variance in amino acid physicochemical properties, and using the proxy descriptors to describe individual amino acids in peptides with known binding affinities.
Multilayer perceptron neural network binding prediction equations
Deriving binding prediction equations for a MHC-I or MHC-II binding partner via neural network modeling by comparing peptide vectors with known binding affinities of the peptides, wherein the principal components of the amino acids are utilized as the input layer of a multilayer perceptron neural network.
Permutation-based binding prediction across MHC binding regions
Applying the neural network to predict the ability of a plurality of peptides of defined length from the target polypeptide to bind to the binding partner, wherein the neural network determines the permuted average binding of a peptide to at least 20 MHC-I or MHC-II binding regions.
Selection of peptides by predicted binding affinity and recombinant synthesis
Selecting one or more peptides from the target molecule having a binding affinity of about greater than 10^6 M^-1 to the binding partner and cloning the selected peptides with an expression vector for recombinantly synthesizing a biomolecule selected from the group consisting of a protein and nucleic acid encoding the one or more peptides.
The claim coverage is directed to generating MHC-I or MHC-II binding prediction equations from experimentally derived amino-acid physicochemical properties using Principal Component Analysis and a multilayer perceptron neural network, predicting binding for peptides of defined length derived from a target polypeptide using permuted average binding over at least 20 MHC binding regions, and selecting peptides above a specified binding-affinity threshold for cloning and recombinant synthesis of encoded protein or nucleic acid.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Application of the framework to influenza A H3N2 antigenic drift by converting a HA1 amino-acid array of 447 viruses into consecutive 9-mer and 15-mer peptides, computing predicted MHC-I and MHC-II binding affinities, and clustering viruses into 23 clusters that largely mirror antibody-defined clusters.
Mapping amino-acid substitutions between antigenic clusters to changes in predicted MHC binding and using defined “epitope mimics” with a principal-component/matrix-distance detection scheme weighted by MHC importance (VIP) to find peptide mimics across pathogen and host proteins.
An example matching IA2 peptides to rotavirus VP7 and an epitope-mapping application comparing predictions to transgenic-mouse human MHC binding data for vaccinia virus (VACV Western Reserve).
Breadth analyses across bacterial/protozoal proteomes by selecting high-affinity MHC-I/MHC-II peptides and B-cell epitope candidates, summarizing peptide counts and conservation across strains, and presenting tables of results.
Interested in licensing this patent?