| Abstrakt: |
Understanding protein behavior depends heavily on where a protein is located within the cell. The successful prediction of protein locations can assess the early diagnosis of a disease and the effectiveness of various drugs. As a result, it is critical to create a protein localization prediction system that is both dependable and accurate. Numerous methods and resources have been employed to obtain insight into protein function over the years. Confocal Microscopy, employed by Human Protein Atlas, is one of them. In this research, the focus has been made on the localization of proteins in 19 subcellular compartments in human cells. The objective is to determine the organelle of the cell that contains the protein. In this research, a deep learning method for categorizing subcellular protein patterns in human cells is presented. The proposed method is primarily based on transfer learning technique using two pre-trained models, VGG-16 and DenseNet121. Due to the highly unbalanced nature of the dataset utilized in this study, class weight optimization is used before training the model. The model's performance is measured using the F1 score, with a final score of 0.75 for VGG-16 and 0.68 for DenseNet121. |
