SRTpred: Machine Learning-Based Prediction of Secretory Proteins Using Amino Acid Composition, Their Order and Similarity Search

SRTpred: Machine Learning-Based Prediction of Secretory Proteins Using Amino Acid Composition, Their Order and Similarity Search SRTpred is a computational web server developed for predicting mammalian secretory proteins from whole protein sequences. The tool predicts secretory proteins irrespective of whether they follow the classical signal peptide-dependent pathway or the non-classical signal peptide-independent pathway. SRTpred uses machine learning models, amino acid composition, dipeptide composition, physicochemical properties, and similarity-search information for prediction. Web Server: http://www.imtech.res.in/raghava/srtpred/ Citation Garg, A., and Raghava, G. P. S. A machine learning based method for the prediction of secretory proteins using amino acid composition, their order and similarity-search. In Silico Biology, 8, 0012, 2008. Article URL: http://www.bioinfo.de/isb/2008/08/0012/main.html About the Research Secretory proteins are proteins that are transported outside the cell or into extracellular space after synthesis. These proteins are important in cell signaling, immune response, extracellular matrix organization, enzyme secretion, and host-pathogen interactions. Classical secretory proteins usually contain an N-terminal signal peptide that directs them to the endoplasmic reticulum and Golgi-dependent secretion pathway. However, some proteins are secreted without a classical N-terminal signal peptide through non-classical or leaderless secretion pathways. Many existing tools mainly depend on N-terminal signal peptide detection and may fail when the correct N-terminus is missing or when the protein follows a non-classical secretion route. SRTpred was developed to predict secretory proteins using the whole protein sequence instead of depending only on the N-terminal signal peptide. Data Compilation: The dataset contained 6975 mammalian protein sequences, including 3321 secretory proteins and 3654 non-secretory proteins. Secretory proteins included both classical and non-classical secreted proteins, while non-secretory proteins were cytoplasmic and/or nuclear proteins. Methodology: SRTpred uses artificial neural network models, support vector machine models, BLAST, PSI-BLAST, and hybrid approaches. Protein features used include physicochemical properties, amino acid composition, dipeptide composition, and similarity-search output.