Deciphering Protein Secretion from the Brain to Cerebrospinal Fluid for Biomarker Discovery

Area of Science:

• Neuroscience
• Biochemistry
• Computational Biology

Background:

• Cerebrospinal fluid (CSF) is crucial for identifying neurological disease biomarkers.
• High protein concentration variability in CSF challenges the detection of low-abundance biomarkers using mass spectrometry.
• Understanding brain protein secretion into CSF is vital for biomarker discovery.

Purpose of the Study:

• To investigate the predictability of brain protein secretion into CSF.
• To develop a computational model for classifying brain proteins based on their CSF secretion potential.
• To enhance the selection of candidate biomarkers for neurological diseases.

Main Methods:

• Combined curated CSF proteome data with Human Protein Atlas brain proteome data.
• Classified brain proteins into CSF-secreted and non-CSF-secreted groups.
• Trained a machine learning model using sequence-based protein features to predict secretion.
• Evaluated model performance using area under the curve (AUC) and tested generalization on novel datasets.

Main Results:

• The machine learning model achieved an AUC of 0.89 for high-confidence CSF proteins.
• Key features for prediction included subcellular localization, signal peptides, and transmembrane regions.
• The model demonstrated good generalization to a larger brain proteome dataset.
• Successfully predicted novel CSF proteins identified through affinity proteomics.

Conclusions:

• Protein secretion into CSF can be predicted using sequence-based features and machine learning.
• The developed classifier aids in understanding brain protein secretion mechanisms.
• This approach supports the selection of more effective CSF biomarker candidates for neurological disorders.