ExoPred

ExoPred predicts protein secretion via exosomes, an unconventional ER/Golgi-independent pathway, to identify proteins—including those lacking signal peptides—secreted in exosomes.

Key Features:

• Dataset Assembly: A curated dataset of 2,992 proteins known to be secreted by exosomes and 2,961 non-exosomal proteins was assembled for model training.
• Model Training: Random forest classifiers were trained on sequence-derived dipeptide composition feature vectors.
• Validation and Performance: Tenfold cross-validation yielded accuracy 69.88% ± 2.08 and AUC 0.76 ± 0.03, and an independent test set achieved accuracy 75.73% and AUC 0.840.

Scientific Applications:

• Proteomics: Aid identification of proteins secreted via unconventional exosomal pathways.
• Cell Biology: Inform studies of exosome-mediated secretion and intercellular communication mechanisms.
• Disease Research: Support investigation of exosome-associated protein secretion in disease progression and therapeutic targeting.

Methodology:

Assembled a labeled dataset (2,992 exosomal, 2,961 non-exosomal), computed dipeptide composition feature vectors from sequences, trained random forest classifiers, and evaluated models by tenfold cross-validation and on an independent test set reporting accuracy and AUC.