Timezone: »

What cleaves? Is proteasomal cleavage prediction reaching a ceiling?
Ingo Ziegler · Bolei Ma · Ercong Nie · Bernd Bischl · David Rügamer · Benjamin Schubert · Emilio Dorigatti
Event URL: https://openreview.net/forum?id=bUyk2atqXqt »

Epitope vaccines are a promising direction to enable precision treatment for cancer, autoimmune diseases, and allergies. Effectively designing such vaccines requires accurate prediction of proteasomal cleavage in order to ensure that the epitopes in the vaccine are presented to T cells by the major histocompatibility complex (MHC). While direct identification of proteasomal cleavage in vitro is cumbersome and low throughput, it is possible to implicitly infer cleavage events from the termini of MHC-presented epitopes, which can be detected in large amounts thanks to recent advances in high-throughput MHC ligandomics. Inferring cleavage events in such a way provides an inherently noisy signal which can be tackled with new developments in the field of deep learning that supposedly make it possible to learn predictors from noisy labels. Inspired by such innovations, we sought to modernize proteasomal cleavage predictors by benchmarking a wide range of recent methods, including LSTMs, transformers, CNNs, and denoising methods, on a recently introduced cleavage dataset. We found that increasing model scale and complexity appeared to deliver limited performance gains, as several methods reached about 88.5\% AUC on C-terminal and 79.5\% AUC on N-terminal cleavage prediction. This suggests that the noise and/or complexity of proteasomal cleavage and the subsequent biological processes of the antigen processing pathway are the major limiting factors for predictive performance rather than the specific modeling approach used. While biological complexity can be tackled by more data and better models, noise and randomness inherently limit the maximum achievable predictive performance. All our datasets and experiments are available at https://anonymous.4open.science/r/cleavage_prediction-E8FD.

Author Information

Ingo Ziegler (Ludwig-Maximilians-Universität München)
Ingo Ziegler

Initially interested in the finance industry, I started my education on the Bachelor's level with a business degree in Finance & Accounting. After various internships in the areas of corporate finance and audit, I become more and more passionate about quantitative areas such as machine learning. Completing a machine learning and data science degree at EPFL's Extension School helped me bridge the gap between my business background and a full undergraduate computer science education. Currently, I'm a Master's student in computational linguistics at LMU Munich, focusing on representation and transfer learning.

Bolei Ma (Ludwig-Maximilians-Universität München)
Ercong Nie (Ludwig-Maximilians-Universität München)

Hi, I am Ercong. Currently, I am a first-year PhD student at Center for Information and Language Processing (CIS), LMU of Munich in computational linguistics / natural language processing. Previously, I obtained my M.Sc. degree in computational linguistics and informatics at LMU of Munich. I finished my bachelor study in German linguistics and Finance at Shanghai Jiao Tong University (SJTU). I also spent one semester at University of Heidelberg for an exchange study in German linguistics. My research interests are Cross-lingual transfer learning, Zero- and few-shot learning, Multilinguality.

Bernd Bischl (LMU)
David Rügamer (LMU Munich)
Benjamin Schubert (Institute of computational biology)
Emilio Dorigatti (Ludwig Maximilian Universitaet Muenchen)

More from the Same Authors