Task-specific model API / hosted only Research use only immunopeptidomics

HLAthena

Presentation predictor trained on the mono-allelic peptidome of 95 cell lines — 186,464 peptides across 95 HLA alleles, fifteen of which had no described motif before — which is the dataset that changed this field more than any architectural idea.

Paper DOI Demo 📖 How it works { } JSON

At a glance

DeveloperBroad Institute / Dana-Farber Cancer Institute (Keskin, Wu, Carr labs)

Released2020-01-13

Licencefree web server; academic use

AvailabilityAPI / hosted only

KindTask-specific model

Regulatory statusResearch use only

What it does

Scores peptide–allele pairs and adds features beyond the peptide: gene expression of the source protein and cleavage context, which is why its predictions track real presentation more closely than binding-only tools.

Notable uses in oncology

The underlying allele panel covers at least one allele in 95% of people worldwide for each of HLA-A, -B and -C — the reason the 'long tail' of alleles shrank at all.

Tasks, data types and cancers

TaskAntigen presentation Peptide–MHC binding Neoantigen prioritisation

Data typeProtein sequence Immunopeptidomics (MS-eluted peptides)Transcriptomics (RNA)

Clinical settingBasic research Cancer vaccine design Immunotherapy

CancerPan-cancer

Inputpeptide + allele + source-gene expression + flanking cleavage context

Outputpresentation score / percentile

Architecture

Familyneural network with peptide, expression and cleavage features

Adding expression is what a peptide-only model structurally cannot do; the price is that you need RNA data for the sample.

Training data

95 mono-allelic B721.221 lines (31 HLA-A, 40 HLA-B, 21 HLA-C, 3 HLA-G), 186,464 unique peptides, median 1,860 per allele.

Training set size186,464 peptides / 95 alleles

InstitutionsBroad Institute, Dana-Farber

Linked datasets

training Mono-allelic HLA class I peptidome (Sarkizova / Abelin) Open download — this model and this dataset come from the same paper

Evaluation

Benchmark / dataset	Metric	Value	External validation	Source
top 0.1% of a 1:999 decoy list	PPV	reported by the authors at 1:999 — a stricter setup than several competitors use	no	Source

How to run

Libraryweb server

APIhttp://hlathena.tools/

No standalone weights; use the web tool or reuse the published peptidome (see the data card) to train your own model.

Regulatory status and intended use

Regulatory statusResearch use only

Intended useResearch use.

Source →

Regulatory status is quoted from the source linked above and can change. Research-use-only models must not be used for clinical decisions.

Limitations and bias

Web-only: not embeddable in an offline pipeline.
Expression features require matched RNA data, so it is not a universal 'peptide in, score out' tool.

Sources

At a glance What it does Tasks, data types and cancers Architecture Training data Evaluation How to run Regulatory status and intended use Limitations and bias Sources

Card provenance

Confidencehigh

Verified2026-09-05 · editorial

Updated2026-09-05

Facts on this card come only from the listed sources; the Hugging Face figures are a cached public snapshot.

Related models

This page is educational — it is not medical advice and does not replace consultation with an oncologist. Diagnostic and treatment decisions are made solely by specialist physicians.