HLA-based methods and compositions and uses thereof

1. A method of treating cancer in a human subject in need thereof comprising:

(a) inputting amino acid sequence information of a set of candidate peptide sequences expressed by cancer cells of a single human subject, using a computer processor, into a trained machine learning immunoprecipitated HLA-peptide presentation prediction model having a peptide presentation prediction value (PPV) of at least 0.35 according to a presentation PPV determination method, to generate a plurality of presentation predictions, wherein each presentation prediction of the plurality of presentation predictions is indicative of a presentation likelihood that a peptide sequence of the set of candidate peptide sequences is presented by an MHC protein of the single human subject; wherein the trained machine learning immunoprecipitated HLA-peptide presentation prediction model comprises:

(i) a plurality of parameters;

and

(ii) a function representing a relation between the amino acid sequence information received as input and the presentation likelihood generated as an output based on the amino acid sequence information and the plurality of parameters;

(b) selecting, based at least on the plurality of presentation predictions, a subset of peptide sequences of the set of candidate peptide sequences to generate a set of selected peptide sequences; and

(c) administering to the single human subject a pharmaceutical composition comprising:

(i) a polypeptide with one or more of the selected peptide sequences,

(ii) a polynucleotide encoding the polypeptide of (i);

(iii) APCs comprising (i) or (ii), or

(iv) T cells comprising a T cell receptor (TCR) specific for an MHC of the single human subject in complex with one or more of the peptide sequences selected in (b).

2. The method of claim 1 , wherein the plurality of parameters are based on mono-allelic training data from training cells expressing (A) a recombinant affinity acceptor tagged MHC protein and (B) an endogenous MHC protein that is different than the MHC protein of the recombinant affinity acceptor peptide tagged MHC protein.

3. The method of claim 2 , wherein the recombinant affinity acceptor tagged MHC protein comprises an affinity acceptor peptide, wherein the affinity acceptor peptide is Biotin Acceptor Protein (BAP).

4. The method of claim 2 , wherein the mono-allelic training data comprises:

(1) a plurality of training peptide sequences, wherein each training peptide sequence of the plurality is associated with the recombinant affinity acceptor peptide tagged MHC protein,

(2) an identity of the recombinant affinity acceptor peptide tagged MHC protein associated with each training peptide sequence of the plurality, and

(3) an observation by mass spectrometry that one or more of the training peptide sequences of the plurality was presented by the recombinant affinity acceptor tagged MHC protein immunoprecipitated from the training cells using an anti-affinity acceptor peptide reagent.

5. The method of claim 4 , wherein the observation is an observation by mass spectrometry that one or more of the training peptide sequences of the plurality was presented by the recombinant affinity acceptor tagged MHC protein immunoprecipitated from lysed training cells using an anti-affinity acceptor peptide reagent.

6. The method of claim 4 , wherein the mono-allelic training data further comprises amino acid physical properties, peptide physical properties, expression level of the source protein of the peptide within a cell, protein stability, protein translation rate, ubiquitination sites, protein degradation rate, translational efficiencies from ribosomal profiling, protein cleavability, protein localization, motifs of host protein that facilitate TAP transport, host protein is subject to autophagy, motifs that favor ribosomal stalling, and/or protein features that favor non-sense mediated degradation (NMD).

7. The method of claim 4 , wherein the training peptide sequences of the plurality comprise sequences of endogenous peptides of the training cells.

8. The method of claim 4 , wherein the mono-allelic training data further comprises sequence information of peptides associated with a disease in a peptide library.

9. The method of claim 4 , wherein the anti-affinity acceptor peptide reagent is biotin.

10. The method of claim 2 , wherein the training cells express:

(a) a first recombinant affinity acceptor tagged MHC protein comprising a first MHC molecule and a first affinity acceptor peptide; and

(b) a second recombinant affinity acceptor tagged MHC protein comprising a second MHC molecule and a second affinity acceptor peptide,

wherein the first MHC molecule and the second MHC molecule are non-identical; and

wherein the first affinity acceptor peptide and the second affinity acceptor peptide are non-identical.

11. The method of claim 2 , wherein the training cells are an antigen presenting cell line.

12. The method of claim 2 ,

wherein the recombinant affinity acceptor tagged MHC protein comprises at least 10 different recombinant affinity acceptor tagged MHC proteins.

13. The method of claim 2 , wherein the training cells express multiple endogenous MHC proteins.

14. The method of claim 1 , wherein the MHC protein of the single human subject is an MHC class II protein.

15. The method of claim 1 , wherein each candidate sequence of the plurality of candidate peptide sequences comprises a cancer specific mutation.

16. The method of claim 1 , wherein the plurality of candidate peptide sequences expressed by cancer cells of a single human subject are identified by comparing whole genome or whole exome sequence information from the cancer cells of the single human subject to whole genome or whole exome sequence information from non-cancer cells of the single human subject, and identifying nucleic acid sequences unique to the cancer cells and not present in the non-cancer cells.

17. The method of claim 1 , further comprising formulating the pharmaceutical composition.

18. The method of claim 1 , wherein the method further comprises training the trained machine learning HLA-peptide presentation prediction model with training data using amino acid sequence information from immunoprecipitated mass spectrometry identified HLA-peptide complexes comprising recombinant affinity acceptor tagged MHC protein from mono-allelic training cells.

19. The method of claim 1 , wherein the presentation PPV determination method comprises inputting amino acid sequence information of a plurality of test peptide sequences into the trained machine learning immunoprecipitated HLA-peptide presentation prediction model to generate a plurality of test presentation predictions, each test presentation prediction indicative of a likelihood that the one or more proteins encoded by an HLA allele can present a given test peptide sequence of the plurality of test peptide sequences,

wherein the plurality of test peptide sequences comprises at least 500 test peptide sequences comprising:

(i) at least one hit peptide sequence identified by mass spectrometry to be presented by an HLA protein expressed in cells, and

(ii) at least 499 decoy peptide sequences contained within a protein encoded by a genome of an organism, wherein the organism and the subject are the same species,

wherein the plurality of test peptide sequences comprises a ratio of 1:499 of the at least one hit peptide sequence to the at least 499 decoy peptide sequences and a top 0.2% of the plurality of test peptide sequences are predicted to be presented by the HLA protein expressed in cells by the trained machine learning HLA-peptide presentation prediction model.

20. The method of claim 19 , wherein (i) the at least one hit peptide sequence comprises at least 10 hit peptide sequences, and (ii) the at least 499 decoy peptide sequences comprise at least 4,990 decoy peptide sequences.

21. The method of claim 1 , wherein the trained machine learning HLA-peptide presentation prediction model has a PPV of more than 0.4 according to the presentation PPV determination method.

22. The method of claim 1 , wherein the trained machine learning HLA-peptide presentation prediction model has a PPV of more than 0.35 at a recall rate of 10% according to the presentation PPV determination method.

23. The method of claim 1 , wherein the trained machine learning HLA-peptide presentation prediction model has a PPV of more than 0.35 at a recall rate of 20% according to the presentation PPV determination method.