Predicting eukaryotic protein secretion without signals.

Area of Science:

• Proteomics
• Bioinformatics
• Molecular Biology

Background:

• Predicting unconventional protein secretion (UPS) in eukaryotes is complex due to limited data and pathway knowledge.
• The SecretomeP method (2004) aimed to predict UPS based on shared properties of secreted proteins.
• The efficacy of SecretomeP and its successors, and the applicability of multi-category predictors for UPS, require critical evaluation.

Purpose of the Study:

• To critically assess the performance of SecretomeP and related methods for predicting unconventional protein secretion in eukaryotes.
• To investigate the potential of multi-category subcellular location predictors for UPS prediction.
• To explore the utility of recent machine learning advancements for developing novel UPS prediction tools.

Main Methods:

• Re-evaluation of the SecretomeP prediction method through a new benchmark study.
• Comparative analysis of SecretomeP against its successors and multi-category predictors.
• Discussion of emerging machine learning approaches for protein secretion prediction.

Main Results:

• A new benchmark indicates SecretomeP's performance is significantly lower than initially estimated.
• The study casts doubt on the hypothesis that all secreted proteins share common properties irrespective of the secretion pathway.
• Existing methods show limitations in accurately predicting unconventional protein secretion in eukaryotes.

Conclusions:

• The performance of SecretomeP is suboptimal, challenging the foundational assumptions of universal secretory pathway features.
• Multi-category subcellular location predictors may not be directly suitable for accurate UPS prediction.
• Recent machine learning developments offer promising avenues for creating more accurate UPS prediction tools that can identify key sequence determinants.