Do Pseudosequences Matter in Neoantigen Prediction?

All pseudosequences and embeddings used in the paper "Do Pseudosequences Matter in Neoantigen Prediction?"

bioRxiv preprint: https://doi.org/10.64898/2025.12.09.693250

All code is freely available at https://github.com/KarchinLab/Do-Pseudoseqs-Matter

--------------------------------------------------------------------------------
FILE NAMES

The following are one-hot encoded pseudosequences, with each row representing an allele and each column representing the amino acid present at a given position. Each entry is either 0 or 1 to indicate whether that residue appears at that position.

Original_BigMHC_Pseudosequences.csv
Original_NetMHCpan-4.1_Pseudosequences.csv
Random_Pseudosequences.csv

BigMHC_Pseudosequences_Length_5.csv
BigMHC_Pseudosequences_Length_10.csv
BigMHC_Pseudosequences_Length_15.csv
BigMHC_Pseudosequences_Length_20.csv
BigMHC_Pseudosequences_Length_25.csv
BigMHC_Pseudosequences_Length_35.csv
BigMHC_Pseudosequences_Length_50.csv
BigMHC_Pseudosequences_Length_80.csv
BigMHC_Pseudosequences_Length_100.csv

The following are embeddings generated by ESM-2 and node2vec. Each row corresponds to an allele, and each column represents a dimension of the embedding space. The entries are floating-point values.

BigMHC_ESM2_Embeddings.csv: ESM-2 embeddings of the original BigMHC pseudosequences
NetMHCpan-4.1_ESM2_Embeddings.csv: ESM-2 embeddings of the original NetMHCpan-4.1 pseudosequences
Annotation_Based_Embeddings.csv: graph-derived allele embeddings constructed using shared P-groups, supertypes, and allele nomenclature

A complete set of files is provided in a single ZIP file:
All_Pseudosequences_and_Embeddings.zip

The following are raw prediction scores described in the paper:
Supplementary_Information_1.xlsx: raw prediction scores for all models
Supplementary_Information_2.xlsx: raw prediction scores for all pseudosequence lengths