Splicing neoepitope prediction is sensitive to methodological differences

Laurie Prélot, Julianne K. David, Andy Lin, André Kahles, Mariia Yurchikova, Reid F. Thompson, Gunnar RätschbioRxiv 2025 · DOI

Abstract

Cancer-specific neoepitopes may arise from abnormal splicing in the transcriptomic landscape (alternative splicing neoepitopes, ASNs), leading to divergent proteins with high potential immunogenicity. Predicting candidate ASNs requires a series of algorithmic and parameter choices, but no previous study has investigated the consistency and interpretability of these choices. Results: We apply two ASN prediction methods with 35 matched parameter sets to generate candidate ASNs for five breast and ovarian cancer samples from The Cancer Genome Atlas (TCGA). We find that: 1) junction 9-mer peptides generated by two similarly designed pipelines differed by, on average, 68.9% and 76.6% in the BRCA and OV cohorts, respectively, and putatively cancer-specific junction 9-mers from the two pipelines diverged further, by an average of 81.5 % in OV and 84.6 % in BRCA; 2) the most lenient filters in the BRCA cohort show the highest divergence, at 97 %; 3) the rate of mass spectrometry validation of ASNs’ protein presence in cells is dominated by the size of the input space; and 4) putatively cancer-specific ASNs found by the intersection of both pipelines can be validated when accounting for false discovery with an average of 1.74 and 284.4 candidates in the BRCA and OV cohorts, respectively. Taken together, these results highlight that ASN identification is fragile and difficult to reproduce across analysis platforms, with limited cross-pipeline overlap and strong dependence on parameter choices.

Introduction

Cancer immunotherapy often relies on vaccines that train the immune system to recognize neoepitopes: tiny protein fragments (peptides) found on the surface of tumor cells but not healthy ones. While most research focuses on neoepitopes caused by DNA mutations, a massive untapped source comes from alternative splicing—errors in how RNA is "cut and pasted" before being translated into protein.

These Alternative Splicing Neoepitopes (ASNs) are promising because they can be highly immunogenic. However, the bioinformatics pipelines used to predict them are complex and unstandardized. This study quantifies the uncertainty inherent in these predictions, demonstrating that the choice of bioinformatics parameters and algorithms—rather than just the underlying biology—drastically dictates which vaccine candidates are identified.

Methodology

The authors performed a head-to-head comparison of two different ASN prediction strategies: a Graph-based Pipeline (GP), which uses a splicing graph to represent all possible exon connections, and a Junction-based Pipeline (JP), which focuses on specific exon-exon junctions inserted into known transcript annotations.

To ensure a fair comparison, the pipelines were aligned using 35 matched sets of parameters across RNA-seq data from five Breast Cancer (BRCA) and five Ovarian Cancer (OV) samples from The Cancer Genome Atlas (TCGA). Candidates were filtered against a normal background of over 9,000 samples from the GTEx project and the UniProt reference proteome. Finally, predictions were cross-referenced with Mass Spectrometry (MS) data from the CPTAC consortium to provide physical evidence of protein presence.

Results: A Landscape of Divergence

The primary finding is a profound lack of consensus between the two pipelines. Even when tuned to use the same logic, the initial pool of 9-mer peptides generated by GP and JP differed by ~69–77%. Once filtered for "cancer-specific" peptides, the divergence jumped to over 80%.

Comparison of filtered junction 9-mer set sizes for GP and JP pipelines across various parameter settings in BRCA (a) and OV (b) cohorts.

The GP pipeline was generally much stricter, producing significantly fewer candidates than the JP pipeline, which was 1.3x to 22.8x more lenient in the BRCA cohort. This architectural difference creates fundamentally different "peptide spaces"; for instance, the JP pipeline can generate peptides spanning three junctions, while the GP pipeline is more restricted.

The Impact of Filtering

The single most influential factor in reducing the candidate list was the GTEx normal tissue filter. If a pipeline is slightly more sensitive to "noise" in normal tissue, it will exclude a massive number of potential tumor targets. The JP pipeline proved extremely sensitive to "recurrence" parameters (how many other patients must share the mutation), while the GP pipeline remained more stable due to its internal graph-based confidence scores.

Filtering cascades for JP and GP pipelines across BRCA and OV cohorts, highlighting the significant reduction in candidates following the GTEx normal tissue filter.

Validation and the Search Space Problem

In Mass Spectrometry validation, the study found that the more candidates searched for, the more were found. However, the rate of validation remained low (around 1–3%), suggesting that many "validated" hits might be a byproduct of a larger search space rather than higher accuracy. The number of candidates confirmed by MS that were found by both pipelines was remarkably small: an average of only 1.74 candidates in BRCA and 284.4 in OV.

Conclusion

ASN identification is currently fragile and difficult to reproduce across analysis platforms. The field should move toward consensus-based prediction—similar to how DNA mutations are now called using multiple algorithms—rather than relying on a single pipeline. A "multi-pipeline intersection" approach is recommended to identify the most robust and defensible vaccine targets for clinical application.

Summary updated

Abstract

Abstract Motivation Cancer-specific neoepitopes may arise from abnormal splicing in the transcriptomic landscape (alternative splicing neoepitopes, ASNs), leading to divergent proteins with high potential immunogenicity. Predicting candidate ASNs requires a series of algorithmic and parameter choices, but no previous study has investigated the consistency and interpretability of these choices. Results We apply two ASN prediction methods with 35 matched parameter sets to generate candidate ASNs for five breast and ovarian cancer samples from The Cancer Genome Atlas (TCGA). We find that: 1) junction 9-mer peptides generated by two similarly designed pipelines differed by, on average, 68.9% and 76.6% in the BRCA and OV cohorts, respectively, and putatively cancer-specific junction 9-mers from the two pipelines diverged further, by an average of 81.5 % in OV and 84.6 % in BRCA; 2) the most lenient filters in the BRCA cohort show the highest divergence, at 97 %; 3) the rate of mass spectrometry validation of ASNs’ protein presence in cells is dominated by the size of the input space; and 4) putatively cancer-specific ASNs found by the intersection of both pipelines can be validated when accounting for false discovery with an average of 1.74 and 284.4 candidates in the BRCA and OV cohorts, respectively. Taken together, these results highlight that ASN identification is fragile and difficult to reproduce across analysis platforms, with limited cross-pipeline overlap and strong dependence on parameter choices. Availability and Implementation Python research code and scripts are available for download at https://github.com/ratschlab/projects2020_ohsu . Contact [email protected] , [email protected]
BibTeX
@misc{prelot2025splicing,
  author = {Laurie Prélot and Julianne K. David and Andy Lin and André Kahles and Mariia Yurchikova and Reid F. Thompson and Gunnar Rätsch},
  title = {{Splicing neoepitope prediction is sensitive to methodological differences}},
  year = {2025},
  doi = {10.1101/2025.09.10.674685},
  howpublished = {bioRxiv},
}
Figure