NCSP-PLM: An ensemble learning framework for predicting non-classical secreted proteins based on protein language...

Area of Science:

• Proteomics and Bioinformatics
• Molecular Biology and Biochemistry

Background:

• Non-classical secreted proteins (NCSPs) are crucial for intercellular communication but lack signal peptides, complicating their identification.
• Experimental methods for NCSP identification are laborious and expensive, necessitating efficient computational approaches.

Purpose of the Study:

• To develop and evaluate an accurate computational method for identifying non-classical secreted proteins (NCSPs).
• To leverage pre-trained protein language models (PLMs) and ensemble deep learning for enhanced NCSP prediction.

Main Methods:

• An ensemble learning framework, NCSP-PLM, was designed using feature embeddings from nine PLMs.
• Deep learning models (MLP, attention, BiLSTM) were fine-tuned and integrated via weighted voting.
• Performance was validated using 5-fold cross-validation and an independent test set.

Main Results:

• NCSP-PLM achieved high performance on benchmark datasets.
• The model demonstrated a sensitivity of 91.18% and specificity of 97.06% on an independent test set.
• Overall accuracy reached 94.12%, significantly outperforming existing state-of-the-art predictors by 7-16%.

Conclusions:

• NCSP-PLM offers a robust and accurate computational tool for identifying NCSPs.
• The findings facilitate deeper understanding of NCSP secretion mechanisms and their roles in cellular processes.
• This approach aids in the large-scale annotation of NCSPs in biological research.