Granted Patent B2

US 12,462,897 · App. 17/281,707 · Granted Nov 4, 2025

Inventors: Brandon Malone (Dossenheim, DE); Kousuke Onoue (Kanagawa, JP); Yoshiko Yoshihara (Kanagawa, JP)

Assignee: NEC CORPORATION

G16B20/20G16B20/30G16B30/10G16B40/20

Abstract

Claims (14)

1 . A method of ranking candidate epitopes derived from neoantigens as targets for personalized immunotherapy, the method comprising:

extracting experimentally-verified properties of epitopes and domain knowledge about the epitopes;

embedding each of the epitopes in a vector space based on the experimentally-verified properties and the domain knowledge;

collecting candidate epitopes based on patient data of a cancer patient;

calculating a set of scores for each of the candidate epitopes, each score in a respective one of the sets for a respective one of the candidate epitopes representing an independent measure of a likelihood of the respective one of candidate epitopes to elicit an immune response in the cancer patient,

wherein each of the sets of scores includes at least a first score indicating a likelihood of human leukocyte antigen (HLA) binding which is determined using HLA alleles which are specific to the cancer patient, and a second score indicating a T-cell response which is predicted using a T-cell receptor (TCR) repertoire which is identified using healthy ribonucleic acid (RNA) sequence data specific to the cancer patient; combining the scores in each of the sets of scores into a single score for each of the candidate epitopes, the single scores for the candidate epitopes in each case reflecting an overall likelihood of eliciting the immune response in the patient;

ranking the candidate epitopes using the single scores and the embeddings for the immunotherapy, wherein the ranking is performed in an order of largest weighted distances in the vector space, the weighted distances in each case being determined based on Euclidean distances in the vector space multiplied by the single score for each respective one of the candidate epitopes such that one of the candidate epitopes with a largest weighted distance from an origin of the vector space is ranked first followed by one of the candidate epitopes having a largest weighted difference from the top-ranked epitope;

producing a neopeptide, a polynucleotide encoding the neopeptide, a vector containing the polynucleotide encoding the neopeptide, a messenger ribonucleic acid (mRNA) encoding the neopeptide, an antigen-presenting cell presenting a complex of the neopeptide and a human leukocyte antigen (HLA) molecule on the surface, an exosome secreted from the antigen-presenting cell, or a combination thereof,

wherein the neopeptide comprises at least one of the ranked candidate epitopes.

2 . The method according to claim 1 , wherein each of the sets of scores further comprises a third score based on tumor RNA-sequence data specific to the cancer patient.

3 . The method according to claim 1 , wherein the embedding is performed using a representation learning embedding framework which uses an affinity graph in which nodes represent the epitopes and edges connect the epitopes which have a similarity measure above a predefined threshold, wherein attributes of the nodes include at least the experimentally-derived properties and the domain knowledge, and wherein an embedding function is learned for each of the attributes to map the attributes to numeric vectors.

4 . The method according to claim 1 , wherein the embedding is performed by direct embedding in which at least the experimentally-derived properties and the domain knowledge are each embedded using numeric vectors which are concatenated together.

5 . The method according to claim 1 , wherein the embeddings include vector representations of biochemical properties of the epitopes.

6 . The method according to claim 1 , wherein the embeddings include vector representations of amino acid sequences of the epitopes.

Assignments (1)

Continuity (2)

References Cited (18)