Research On Protein Subcellular Localization Prediction Based On Transductive Learning

As an important research field in proteomics, protein subcellular localization is closely related to protein function, metabolic way and other biological processes, and plays an important role in drug found, disease diagnosis, etc. Subcellular localization prediction method based on machine learning is a kind of efficient prediction method, but the machine learning method requires a lot of training data to improve prediction accuracy, but the experimental data used for training is not very much in protein subcellular localization prediction, multi-location protein deficiency phenomenon is especially obvious. This paper proposes a subcellular localization prediction method based on transductive learning, which can improve prediction performance of protein subcellular localization in case of lacking of multi-location protein. This method can make full use of the sample feature information of testing dataset and infer directly on the testing set by calculating the similarity relationship between all samples, then get subcellular localization information of testing dataset finally.We conducted experiments on the dataset of human, plant and virus to verify the effectiveness of our algorithm. The results compared with the current correlation algorithm show that our method can effectively improve the predict performance of multi-location protein without reducing the overall prediction accuracy of protein subcellular localization.