pepsickle rapidly and accurately predicts proteasomal cleavage sites for improved neoantigen identification

Weeder, Benjamin R.; Wood, M. C.; Li, Ellysia; Nellore, Abhinav; Thompson, Reid F.

doi:10.1093/bioinformatics/btab628

Cited by 6 publications

(15 citation statements)

References 79 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…NetCleave [Amengual-Rigo and Guallar, 2021] extended cleavage prediction to MHC-II and H2 ligands and returned to neural network models, while also using VHSE amino acid descriptors. Pepsickle [Weeder et al, 2021] collected a larger dataset of in vitro cleavage sites and showed that using predicted C-terminal cleavage probabilities helped isolating immune-responsive epitopes, thus improving EV design. Finally, Dorigatti et al [2022] proposed to treat decoy samples as unlabeled, rather than negatives, and used specific techniques that do not require labeled negatives to learn a classifier [Bekker and Davis, 2020].…”

Section: Evolution Of Proteasomal Cleavage Predictorsmentioning

confidence: 99%

“…2). Even though such negative samples are not entirely reliable, the growing availability of this kind of data [Vita et al, 2018] spurred continuous development and improvement of proteasomal cleavage predictors [Keşmir et al, 2002, Kuttler et al, 2000, Dönnes and Kohlbacher, 2005, Nielsen et al, 2005b], which have been recently revised in light of new innovations in the deep learning field [Amengual-Rigo and Guallar, 2021, Dorigatti et al, 2022, Weeder et al, 2021].…”

Section: Evolution Of Proteasomal Cleavage Predictorsmentioning

confidence: 99%

“…As baseline models, we considered a logistic regression classifier with L2 regularization, which is essentially equivalent to ProteaSMM, tuned via 10-folds cross-validation, and a SVM model fitted using 15 000 samples to computational limitations, optimizing its hyperparameters via three-folds cross-validation. As for published predictors, we bench-marked against NetCleave [Amengual-Rigo and Guallar, 2021], Pepsickle [Weeder et al, 2021], and PUUPL [Dorigatti et al, 2022], retraining them on our dataset of both N- and C-terminals, as well as comparing a retrained version of PCM [Dönnes and Kohlbacher, 2005] with the published pretrained model. We also considered PCPS (Model 1) [Diez-Rivero et al, 2010], NetChop 20S and NetChop C-term [Nielsen et al, 2005a], however, we could not re-train these due to the unavailability of the source code.…”

Section: Benchmarking Of Proteasomal Cleavage Prediction Strategiesmentioning

confidence: 99%

See 2 more Smart Citations

Proteasomal cleavage prediction: state-of-the-art and future directions

Ziegler

Bolei

Bischl

et al. 2023

Preprint

View full text Add to dashboard Cite

Epitope vaccines are a promising approach for precision treatment of pathogens, cancer, autoimmune diseases, and allergies. Effectively designing such vaccines requires accurate proteasomal cleavage prediction to ensure that the epitopes included in the vaccine trigger an immune response. The performance of proteasomal cleavage predictors has been steadily improving over the past decades owing to increasing data availability and methodological advances. In this review, we summarize the current proteasomal cleavage prediction landscape and, in light of recent progress in the field of deep learning, develop and compare a wide range of recent architectures and techniques, including long short-term memory (LSTM), transformers, and convolutional neural networks (CNN), as well as four different denoising techniques. All open-source cleavage predictors re-trained on our dataset performed within two AUC percentage points. Our comprehensive deep learning architecture benchmark improved performance by 0.4 AUC percentage points, while closed-source predictors performed considerably worse. We found that a wide range of architectures and training regimes all result in very similar performance, suggesting that the specific modeling approach employed has a limited impact on predictive performance compared to the specifics of the dataset employed. We speculate that the noise and implicit nature of data acquisition techniques used for training proteasomal cleavage prediction models and the complexity of biological processes of the antigen processing pathway are the major limiting factors. While biological complexity can be tackled by more data and, to a lesser extent, better models, noise and randomness inherently limit the maximum achievable predictive performance.

show abstract

Section: Evolution Of Proteasomal Cleavage Predictorsmentioning

confidence: 99%

Section: Evolution Of Proteasomal Cleavage Predictorsmentioning

confidence: 99%

Section: Benchmarking Of Proteasomal Cleavage Prediction Strategiesmentioning

confidence: 99%

See 1 more Smart Citation

Proteasomal cleavage prediction: state-of-the-art and future directions

Ziegler

Bolei

Bischl

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Consequently, several approaches are undertaken to allow for the prediction of the immunopeptidome in silico to overcome the costs and time demands of immunopeptidomics. One such tool is the open-source software Pepsickle (version 0.1.2), which allows one to calculate the cleavage probability of proteins by the c20S or the i20S [ 29 ]. Pepsickle utilizes a deep ensemble learning algorithm together with a set of experimental proteasomal training data.…”

Section: Introductionmentioning

confidence: 99%

“…Furthermore, the tool considers the upstream and downstream amino acid context at each cleavage site for its calculation. Pepsickle allows for a fast and reliable calculation of the estimated peptide pool generated by the proteasome that will predominantly be loaded onto MHC class I [ 29 ].…”

Section: Introductionmentioning

confidence: 99%

A Lysine Residue at the C-Terminus of MHC Class I Ligands Correlates with Low C-Terminal Proteasomal Cleavage Probability

Schmalen,

Kammerl,

Meiners

et al. 2023

Biomolecules

View full text Add to dashboard Cite

The majority of peptides presented by MHC class I result from proteasomal protein turnover. The specialized immunoproteasome, which is induced during inflammation, plays a major role in antigenic peptide generation. However, other cellular proteases can, either alone or together with the proteasome, contribute peptides to MHC class I loading non-canonically. We used an immunopeptidomics workflow combined with prediction software for proteasomal cleavage probabilities to analyze how inflammatory conditions affect the proteasomal processing of immune epitopes presented by A549 cells. The treatment of A549 cells with IFNγ enhanced the proteasomal cleavage probability of MHC class I ligands for both the constitutive proteasome and the immunoproteasome. Furthermore, IFNγ alters the contribution of the different HLA allotypes to the immunopeptidome. When we calculated the HLA allotype-specific proteasomal cleavage probabilities for MHC class I ligands, the peptides presented by HLA-A*30:01 showed characteristics hinting at a reduced C-terminal proteasomal cleavage probability independently of the type of proteasome. This was confirmed by HLA-A*30:01 ligands from the immune epitope database, which also showed this effect. Furthermore, two additional HLA allotypes, namely, HLA-A*03:01 and HLA-A*11:01, presented peptides with a markedly reduced C-terminal proteasomal cleavage probability. The peptides eluted from all three HLA allotypes shared a peptide binding motif with a C-terminal lysine residue, suggesting that this lysine residue impairs proteasome-dependent HLA ligand production and might, in turn, favor peptide generation by other cellular proteases.

show abstract