Method and system for binding affinity prediction and method of generating a candidate protein-binding peptide

1 . A computer-implemented method of predicting a binding affinity of a query binder peptide to a query target major histocompatibility complex (MHC) molecule from a subject with cancer, the method comprising:

obtaining, with at least one processor, a first amino acid sequence corresponding to the query binder peptide and a second amino acid sequence with defined residue numbering corresponding to the query target MHC molecule, wherein the first amino acid sequence is encoded by a DNA sequence of a tumor biopsy from the subject;

accessing, with the at least one processor for the subject with cancer, a reference data store of reference binder-target pairs comprising respective paired reference binder amino acid sequences and corresponding reference target amino acid sequences, each reference binder-target pair having an associated known binding value;

generating, with the at least one processor, a representation of the first amino acid sequence as a set of query binder subsequences which collectively span across the first amino acid sequence, each query binder subsequence comprising one or more amino acid residues at respective positions along the first amino acid sequence and having a length less than the length of the first amino acid sequence, wherein at least one of the query binder subsequences comprises two or more amino acid residues;

for each query binder subsequence of the set of query binder subsequences, determining contact positions of contact amino acid residues in the second amino acid sequence from a contact point map comprising the residue numbering and assembling a corresponding query target subsequence from the contact amino acid residues at the contact positions in the contact point map, to thereby generate query binder-target subsequence pairs, such that each query target subsequence represents MHC amino acid residues in proximity to each amino acid residue of the corresponding query binder subsequence according to the contact point map;

generating, with the at least one processor, from the reference binder-target pairs, a reference data set comprising a plurality of sets of reference binder-target subsequence pairs, each reference binder-target subsequence pair in each set of reference binder-target subsequence pairs having a corresponding query binder subsequence in the set of query binder subsequences, each reference binder-target subsequence pair comprising: a reference binder subsequence comprising amino acid residues of the respective reference binder amino acid sequence at positions corresponding to those positions of the corresponding query binder subsequence, and a reference target subsequence comprising amino acid residues of the respective reference target sequence at the contact positions derived from the contact point map;

assigning each reference binder-target subsequence pair a reference binding value based on the known binding value of the reference binder-target pair from which it was generated, such that each reference binder-target pair is decomposed into a set of subsequences corresponding to each of the query binder subsequences and the corresponding query target subsequence, with each set of subsequences assigned the known binding value of the reference binder-target pair;

performing, with the at least one processor, at least one similarity operation on the respective query binder-target subsequence pairs and the sets of reference binder-target subsequence pairs to generate a plurality of similarity scores for each query binder-target subsequence pair, such that the similarity scores for each query binder-target subsequence pair represent a contribution to binding affinity from subsequences in a set of reference binder-target subsequence pairs that are similar to the respective query binder-target subsequence, wherein the at least one similarity operation generates similarity scores by generating a similarity score for a comparison between a query binder subsequence and reference binder subsequence, wherein the at least one similarity operation comprises: generating a bipartite graph comprising a first set of nodes and a second set of nodes, the first set of nodes containing only binder subsequences and the second set of nodes containing only target subsequences, edge weights of the bipartite graph being equal to associated known binding values; and determining a monopartite projection of the bipartite graph for the first set of nodes and/or the second set of nodes, wherein similarity scores are edge weights of the monopartite projection for the first set of nodes and/or the second set of nodes;

computing, with the at least one processor, the binding affinity for the query binder peptide to the query target MHC molecule as a weighted combination of the reference binding values of the sets of reference binder-target subsequence pairs, wherein weights of the weighted combination are based on the plurality of similarity scores for each query binder-target subsequence pair;

determining, with the at least one processor, that the binding affinity for the query binder peptide to the query target MHC molecule satisfies a binding threshold;

selecting, with the at least one processor, responsive to determining that the binding affinity satisfies the binding threshold, the query binder peptide as a candidate peptide for creating a vaccine;

sending, to an automated peptide synthesis device, a peptide sequence corresponding to the candidate peptide to cause the automated peptide synthesis device to create the vaccine including the candidate peptide; and

administering to the subject the vaccine including the candidate peptide.

2 . The computer-implemented method according to claim 1 , wherein the at least one similarity operation generates similarity scores by generating a second similarity score for a comparison between a query target subsequence and reference target subsequence, and combines the similarity score for the comparison between the query binder subsequence and the reference binder subsequence and the second similarity score.

3 . The computer-implemented method according to claim 2 , wherein the similarity score for the comparison between the query binder subsequence and the reference binder subsequence is given a non-zero value in the case of an exact match, and a zero value otherwise and/or wherein the second similarity score is given a non-zero value in the case of an exact match, and a zero value otherwise.

4 . The computer-implemented method according to claim 1 , wherein the at least one similarity operation comprises a sequence alignment between a query binder subsequence and reference binder subsequence, and/or a sequence alignment between a query target subsequence and reference target subsequence.

5 . The computer-implemented method according to claim 1 , wherein a similarity score for a pair of nodes of the first set is computed by determining a set of common nodes of the second set to which both nodes of the pair of nodes of the first set are connected; and computing a linear correlation between the corresponding edge weights of the bipartite graph.

6 . The computer-implemented method according to claim 1 , wherein a similarity score for a pair of nodes of the second set is computed by determining a set of common nodes of the first set to which both nodes of the pair of nodes of the second set are connected; and computing a linear correlation between the corresponding edge weights of the bipartite graph.

7 . The computer-implemented method according to claim 1 , wherein the step of determining contact positions further comprises simulating a virtual query binder subsequence for use in the contact point map based on the query binder subsequence, wherein the virtual query binder subsequence has a different length to the query binder subsequence, and/or simulating a set of contact points for use in the contact point map.

8 . The computer-implemented method according to claim 1 , wherein a maximum length of a query binder subsequence is L, L>1, and the set of one or more query binder subsequences comprises all possible subsequences of the first amino acid sequence of length between 1 and L.

9 . The computer-implemented method according to claim 1 , wherein the at least one similarity operation comprises weighting each reference binder-target subsequence pair according to the amino acid positions of the respective reference binder subsequence.

10 . The computer-implemented method according to claim 1 , wherein the second amino acid sequence is a human leukocyte antigen (HLA) protein sequence.

11 . The computer-implemented method according to claim 1 , wherein computing the binding affinity comprises computing a product of a reference data matrix, a query data transpose matrix, and a vector of the reference binding values, wherein entries of the reference data matrix are indicative of presence or absence of reference subsequences in respective reference binder sequences and/or reference target sequences, and entries of the query data transpose matrix are indicative of presence or absence of reference subsequences in respective query binder sequences and/or query target sequences; and wherein the entries of the reference data matrix and the entries of the query data matrix are weighted according to the plurality of similarity scores for each query binder-target subsequence pair.

12 . The computer-implemented method of claim 1 , wherein the obtaining step comprises obtaining a plurality of amino acid sequences corresponding to a plurality of query binder peptides, the first amino acid sequence being one of the plurality of amino acid sequences,

wherein the computing step further comprises computing, for each peptide of the plurality of query binder peptides, a respective binding affinity to the query target MHC molecule, and

wherein selecting the query binder peptide further comprises selecting one or more candidate peptides from the plurality of query binder peptides based on the respective computed binding affinity.

13 . The computer-implemented method of claim 12 , further comprising a method of making the vaccine which comprises (i) synthesizing the one or more candidate peptides using the automated peptide synthesis device in accordance with the peptide sequence or (ii) encoding the one or more candidate peptides into a corresponding DNA or RNA sequence and optionally incorporating the DNA or RNA sequence into a genome of a bacterial or viral delivery system.

14 . The computer-implemented method of claim 1 , wherein the binding binder threshold comprises an IC 50 value of less than or equal to 500 nM.