Abstract
Signal peptides are short N-terminal amino acid sequences that direct proteins through cellular secretion pathways and play an essential role in protein localization and function. Computational tools such as SignalP are widely used to predict signal peptides. However, prediction accuracy for archaeal proteins remains lower than for bacterial and eukaryotic proteins, likely because relatively few archaeal proteins were available for model training. The goal of this study was to determine whether expanding the archaeal training data used by SignalP 6.0 could improve archaeal signal peptide prediction. Archaeal proteomics datasets from five archaeal species were analyzed to identify proteins with experimentally supported signal peptides. Experimental evidence obtained from N-terminal peptides and semi-enzymatic cleavage sites was combined with computational predictions generated by SignalP 6.0 and DeepTMHMM to curate archaeal proteins for model training. These proteins were added to the original SignalP 6.0 training dataset to create two expanded training datasets. One dataset was supplemented with curated Haloferax volcanii proteins, while the other was supplemented with curated proteins from all five archaeal species. A separate shared held-out test set was also generated from the curated archaeal proteins to evaluate the retrained models. Retraining with the expanded archaeal datasets maintained similar overall model performance while improving prediction performance for archaeal proteins. The greatest improvement was observed when proteins from all five archaeal species were included in the training dataset, resulting in an increase in archaeal Matthews Correlation Coefficient (MCC) from 0.715 for the original model to 0.739. Overall MCC remained comparable across all three models, indicating that the addition of archaeal training data improved archaeal prediction performance without substantially affecting prediction performance for the other organism groups. Overall, this study shows that expanding archaeal training data with experimentally supported proteins can improve archaeal signal peptide prediction while maintaining strong overall model performance. Improving signal peptide prediction will enable more accurate identification of secreted archaeal protein, supporting future studies of archaeal protein secretion and the biological processes that depend on these pathways.
Publication Date
8-4-2026
Document Type
Thesis
Student Type
Graduate
Degree Name
Bioinformatics (MS)
College
College of Science
Advisor
Stefan Schulze
Advisor/Committee Member
Julie Thomas
Advisor/Committee Member
Dukka KC
Recommended Citation
Kaur, Anjali, "Archaeal Signal Peptide Prediction Through Integration of Proteomics and Machine Learning" (2026). Thesis. Rochester Institute of Technology. Accessed from
https://repository.rit.edu/theses/12754
Campus
RIT – Main Campus
