Deep Learning Approaches for Classification Subcellular Protein Patterns in Human Cells

Deep learning is an approach that has found applications in many areas of science. In the present work, these learning methods are applied in the field of medicine, namely, to identify a class of proteins in an extremely unbalanced sample. Proteins are the most important component of human cells. Research of these substances helps a person to know himself, strengthen his health and prolong life. However, determining the exact amount and class of proteins in a human cell by hand is quite difficult. That is why it is necessary to create an algorithm capable of helping humanity in solving this problem by determining 28 classes of human proteins. Thus, the goal of our research is to create a protein recognition algorithm that is capable of doing it better than humans. With the help of depth learning algorithms, we were able to develop an approach to solving the problem. According to the results of the study, we obtained a new method, applied it to real data and took the 9th place from more than 2 thousand teams from around the world on the competitive platform Kaggle. Moreover, we have achieved the primary goal and were able to achieve a better quality of protein recognition than ensemble of experts.