Weakly-supervised individual human cell classification

Abstract

Proteins have an essential role in all cellular processes and greatly determine their functional properties. With the development of microscopy, we have the ability to study localization patterns of proteins within a cell group. Differences in the location among cell populations are called single-cell variability (SCV). SCV accounts for a difference in reactions between cells. Studying this connection may aid the development of treatment for cancer [23]. To this end, we develop a cell classifier trained to determine the correspondence of protein location to cell organelles in a single cell. A classifier is trained with image-level labels, so the training process is weakly supervised.