SURFY

SURFY predicts the human surface proteome (surfaceome) to identify and annotate cell-surface proteins relevant for drug targeting and cellular phenotype analysis.

Key Features:

• Random forest classifier: Uses a random forest classifier trained on 131 features per protein and topological domain.
• Training data: Training set comprised high-confidence cell-surface proteins from the Cell Surface Protein Atlas (CSPA).
• Predicted surfaceome: Predicts a human surfaceome of 2,886 proteins.
• Performance: Reports an accuracy of 93.5% for surfaceome prediction.
• Overlap with known classes: Predicted surfaceome shows significant overlap with known cell-surface protein classes, including receptors.
• Expression variability: Reveals cell-type–specific expression variability with 543–1,100 surfaceome genes in cancer cell lines and up to 1,700 in embryonic stem cells and derivatives.
• Drug relevance: Notes that cell-surface proteins account for 66% of approved drug targets in DrugBank.
• Transmembrane protein context: Compares predictions to approximately 5,000 predicted human transmembrane proteins, of which only a fraction are experimentally confirmed at the plasma membrane.
• Multiomics integration: Enables integration of surfaceome predictions with multiomics datasets for downstream analyses.

Scientific Applications:

• Drug target identification: Supports identification and prioritization of cell-surface drug targets.
• Cancer biology: Enables analysis of surfaceome changes across cancer cell lines and identification of cancer-associated surface proteins.
• Stem cell research: Facilitates characterization of surfaceome composition in embryonic stem cells and their derivatives.
• Biomarker discovery: Supports discovery of cellular phenotypes and potential biomarkers through surfaceome analysis.
• Nanoscale organization studies: Aids in elucidating the nanoscale organization of the surfaceome.

Methodology:

Machine learning using a random forest classifier trained on 131 features per protein and topological domain with high-confidence cell-surface proteins from the Cell Surface Protein Atlas (CSPA) as the training set.