LAUSR

Splicing dysregulation is a common feature of cancer, and may produce tumor specific proteins that have altered function or antigenicity. While many tools are available to detect splicing in RNAseq data, robust methods for translating the effects of alternative splicing to altered protein sequences are lacking. In addition, statistical approaches for detecting splicing defects in a tumor compared to normal tissues are only now being considered by bioinformaticians. We have developed a protein sequence generation method and splicing outlier detection strategy and validated this pipeline using a dataset with matched RNAseq and mass spectrometry data from normal tissues. Splice events were detected using SplAdder and our protein sequence generation and annotation tools were implemented in python. We developed a novel approach to outlier detection using a beta binomial model, where splice isoform read counts are modeled with a binomial distribution, and the expected proportion of each splice isoform (‘percent spliced in9 or ‘PSI9) is modeled as the beta distribution based on the PSI observed in a reference dataset. In a normal tissue validation dataset, we detected isoform-specific peptides for 77,369 splice events, including 2,171 that encode novel protein sequences. Both isoforms were detected for 1,476 of these events, and 271 events showed tissue exclusive detection patterns. To evaluate our proposed outlier detection method, we used a set of gastrointestinal (GI) tissues as a reference set and looked for outlier splice events in normal brain, tonsil, and ovarian tissues. We identified examples of peptides where one isoform was exclusively detected in the GI tissues and the other isoform was detected in one of the other tissues. Our outlier detection method was able to predict these events from the RNAseq data with better specificity compared to thresholds set using the distribution of PSIs. We also implemented a differential splicing test using the beta binomial model and benchmarked it against the more commonly used negative binomial model using the normal tissue dataset, and found our proposed method improved sensitivity and specificity. We then applied our splicing outlier detection pipeline to tumor data from TCGA using GTEx as a reference set, and were able to identify tumor specific splice events that generated novel protein sequences. Our splicing pipeline enables the identification of tumor-specific isoforms which may be candidate targets for immunotherapies. Citation Format: Rebecca F. Halperin, Apurva Hegde, Jessica Lang, Elizabeth Raupach, Patrick Pirrotte, Nicholas Schork. A proteomics validated pipeline for detection of differential and tumor-specific splice events [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr LB-244.