MHCflurry: Open-Source Class I MHC Binding Affinity Prediction

O’Donnell, Timothy J.; Rubinsteyn, Alex; Bonsack, Maria; Riemer, Angelika B.; Laserson, Uri; Hammerbacher, Jeff

doi:10.1016/j.cels.2018.05.014

Cited by 346 publications

(374 citation statements)

References 15 publications

Supporting

Mentioning

370

Contrasting

Order By: Relevance

“…For most benchmarks in this work, we used the standard upper limit of 50,000 nM, so that predicted binding affinity was = (0,1 − /01 ( 50)). For the Bonsack et al dataset (12), the upper limit was changed to 100,000nM because in their experiments, as described in O'Donnell et al (23), binders were defined as peptides with IC50<100,000nM. As binding affinity was determined based on in vitro HLA binding-competition vs. a known strong binder (reported IC50 <50nM) experimental IC50 values were in µM range.…”

Section: Supplementary Informationmentioning

confidence: 99%

“…More recently, advances in immunopeptidomics technologies have enabled identification of thousands of naturally presented MHC bound peptides (HLAp) from cancer patient samples and cell lines (27) (28) (24). The potential to improve neoantigen predictors by integrating binding affinity and HLAp data (24) has motivated new hybrid approaches (19,23). Despite these advances, most methods predict large numbers of peptides as candidate neoantigens, of which only a few are actually immunogenic in patients 4 (16,24,29).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

Shao

Bhattacharya

Huang

et al. 2019

Preprint

View full text Add to dashboard Cite

Running title: High-throughput prediction of neoantigens with MHCnuggets Potential Conflicts of Interest: V.A receives research funding from Bristol-Myers Abstract Computational prediction of binding between neoantigen peptides and major histocompatibility complex (MHC) proteins is an emerging biomarker for predicting patient response to cancer immunotherapy. Current neoantigen predictors focus on in silico estimation of MHC binding affinity and are limited by low positive predictive value for actual peptide presentation, inadequate support for rare MHC alleles and poor scalability to high-throughput data sets. To address these limitations, we developed MHCnuggets, a deep neural network method to predict peptide-MHC binding. MHCnuggets is the only method to handle binding prediction for common or rare alleles of MHC Class I or II, with a single neural network architecture. Using a long shortterm memory network (LSTM), MHCnuggets accepts peptides of variable length and is capable of faster performance than other methods. When compared to methods that integrate binding affinity and HLAp data from mass spectrometry, MHCnuggets yields a fourfold increase in positive predictive value on independent MHC-bound peptide (HLAp) data. We applied MHCnuggets to 26 cancer types in TCGA, processing 52.6 million allele-peptide comparisons in under 2.3 hours, yielding 103,587 candidate immunogenic missense mutations (IMMs). IMM hotspots occurred in 36 genes, including 22 driver genes. Predicted IMM load was significantly associated with increased immune cell infiltration (p<2e-16) including CD8+ T cells. Notably, only 0.15% of predicted immunogenic missense mutations were observed in >2 patients, with 65% of these derived from driver mutations. Our results provide a new method for neoantigen prediction with high performance characteristics and demonstrate its utility in large data sets across human cancers. SynopsisWe developed a new in silico predictor of Major Histocompatibility Complex (MHC) ligand binding and demonstrated its utility to assess potential neoantigens and immunogenic missense mutations (IMMs) in 6613 TCGA patients.

show abstract

Section: Supplementary Informationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

Shao

Bhattacharya

Huang

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Similar to [12], the training procedure included 1-cycle learning rate scheduling [15] and discriminative learning rates [13] during finetuning. Target variables for the regression model were log-transformed half-maximal inhibitory concentration (IC 50 )-values and a modified MSE loss function [8] that allows to incorporate qualitative data.…”

Section: A Usmpep: Universal Sequence Models For Peptide Binding Prementioning

confidence: 99%

“…The training data of the tools reported in the literature vary in size and compilation. We trained our models on data provided by [8] and refer to this dataset as MHCFlurry18. It is assembled from an IEDB snapshot of December 2017 and the Kim14 dataset.…”

Section: B Mhc Binding Prediction Datasetsmentioning

confidence: 99%

“…3) Kim14 Dataset: As final benchmark dataset for MHC class I prediction, we consider the Kim14 dataset that is interesting for a number of reasons: In order to investigate how the predictive power of our approach depends on the size of the training data set, we trained and tested our model on the Kim14 BD2009 and Blind data. The authors of [8] kindly provided us with the Blind predictions of their tool trained on BD2009, which allow for a direct comparison with a state-ofthe-art tool. Corresponding training routines are by now also available in the code repository accompanying [8].…”

Section: ) Iedb16 Datasetmentioning

confidence: 99%

See 1 more Smart Citation

USMPep: Universal Sequence Models for Major Histocompatibility Complex Binding Affinity Prediction

Vielhaben

Wenzel

Samek

et al. 2019

Preprint

View full text Add to dashboard Cite

Background: Immunotherapy is a promising route towards personalized cancer treatment. A key algorithmic challenge in this process is to decide if a given peptide (neoepitope) binds with the major histocompatibility complex (MHC). This is an active area of research and there are many MHC binding prediction algorithms that can predict the MHC binding affinity for a given peptide to a high degree of accuracy. However, most of the state-of-the-art approaches make use of complicated training and model selection procedures, are restricted to peptides of a certain length and/or rely on heuristics. Results: We put forward USMPep, a simple recurrent neural network that reaches state-of-the-art approaches on MHC class I binding prediction with a single, generic architecture and even a single set of hyperparameters both on IEDB benchmark datasets and on the very recent HPV dataset. Moreover, the algorithm is competitive for a single model trained from scratch, while ensembling multiple regressors and language model pretraining can still slightly improve the performance. The direct application of the approach to MHC class II binding prediction shows a solid performance despite of limited training data. Conclusions: We demonstrate that competitive performance in MHC binding affinity prediction can be reached with a standard architecture and training procedure without relying on any heuristics.Index Terms-major histocompatibility complex, binding affinity prediction, peptide data, recurrent neural networks, language modeling All authors are with Fraunhofer Heinrich Hertz Institute,

show abstract

Cancer Immunotherapies Ignited by a Thorough Machine Learning‐Based Selection of Neoantigens

Jurczak,

Druchok

2024

Advanced Biology

View full text Add to dashboard Cite

Identification of neoantigens, derived from somatic DNA alterations, emerges as a promising strategy for cancer immunotherapies. However, not all somatic mutations result in immunogenicity, hence, efficient tools to predict the immunogenicity of neoepitopes are needed. A pipeline is presented that provides a comprehensive solution for the identification of neoepitopes based on genomic sequencing data. The pipeline consists of a data pre‐processing step and three machine learning predictive steps. The pre‐processing step analyzes genomic data for different types of alterations, produces a list of all possible antigens, and determines the human leukocyte antigen (HLA) type and T‐cell receptor (TCR) repertoire. The first predictive step performs a classification into antigens and neoantigens, selecting neoantigens for further consideration. The next step predicts the strength of binding between neoantigens and available major histocompatibility complexes of class I (MHC‐I). The third step is engaged to predict the likelihood of inducing an immune response. Neoepitopes satisfying all three predictive stages are assumed to be potent candidates to ensure immunogenicity. The predictive pipeline is used in two regimes: selecting neoantigens from patients' sequencing data and generating novel neoantigen candidates. Two different techniques — Monte Carlo and Reinforcement Learning – are implemented to facilitate the generative regime.

show abstract

MHCflurry: Open-Source Class I MHC Binding Affinity Prediction

Cited by 346 publications

References 15 publications

High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

USMPep: Universal Sequence Models for Major Histocompatibility Complex Binding Affinity Prediction

Cancer Immunotherapies Ignited by a Thorough Machine Learning‐Based Selection of Neoantigens

Contact Info

Product

Resources

About