Protein subcellular localization in human and hamster cell lines: Employing local ternary patterns of fluorescence microscopy images

• LTPs exploit small variations in intensities of Human and Hamster protein images.
• SMOTE oversampling is utilized to increase the minority class samples.
• SVM shows significance performance improvement for balanced data.
• mRMR is not required for the performance improvement of LTPs.
• A web server is available online at http://111.68.99.218/Protein_SubLoc.

Discriminative feature extraction technique is always required for the development of accurate and efficient prediction systems for protein subcellular localization so that effective drugs can be developed. In this work, we showed that Local Ternary Patterns (LTPs) effectively exploit small variations in pixel intensities; present in fluorescence microscopy based protein images of human and hamster cell lines. Further, Synthetic Minority Oversampling Technique is applied to balance the feature space for the classification stage. We observed that LTPs coupled with data balancing technique could enable a classifier, in this case support vector machine, to yield good performance. The proposed ensemble based prediction system, using 10-fold cross-validation, has yielded better performance compared to existing techniques in predicting various subcellular compartments for both 2D HeLa and CHO datasets. The proposed predictor is available online at: http://111.68.99.218/Protein_SubLoc/, which is freely accessible to the public.

Figure optionsDownload as PowerPoint slide