Predicting Protein Submitochondrial Locations Using a K-Nearest Neighbors Method Based on the Bit-Score Weighted Euclidean Distance

摘要

Mitochondria are essential subcellular organelles found in eukaryotic cells. Knowing information on a protein's subcellular or sub-subcellular location provides in-depth insights about the microenvironment where it interacts with other molecules and is crucial for inferring the protein's function. Therefore, it is important to predict the submitochondrial localization of mitochondri-al proteins. In this study, we introduced MitoBSKnn, a K-nearest neighbor method based on a bit-score weighted Euclidean distance, which is calculated from an extended version of pseudo-amino acid composition. We then improved the method by applying a heuristic feature selection process. Using the selected features, the final method achieved a 93% overall accuracy on the benchmarking dataset, which is higher than or comparable to other state-of-art methods. On a larger recently curated dataset, the method also achieved a consistent performance of 90% overall accuracy.