Learning unsupervised feature representations for single cell microscopy images with paired cell inpainting.

Alex X Lu,Sam Cooper,Oren Z Kraus,Alan M Moses,Juan Caicedo

doi:10.1371/journal.pcbi.1007348

Abstract

Cellular microscopy images contain rich insights about biology. To extract this information, researchers use features, or measurements of the patterns of interest in the images. Here, we introduce a convolutional neural network (CNN) to automatically design features for fluorescence microscopy. We use a self-supervised method to learn feature representations of single cells in microscopy images without labelled training data. We train CNNs on a simple task that leverages the inherent structure of microscopy images and controls for variation in cell morphology and imaging: given one cell from an image, the CNN is asked to predict the fluorescence pattern in a second different cell from the same image. We show that our method learns high-quality features that describe protein expression patterns in single cells both yeast and human microscopy datasets. Moreover, we demonstrate that our features are useful for exploratory biological analysis, by capturing high-resolution cellular components in a proteome-wide cluster analysis of human proteins, and by quantifying multi-localized proteins and single-cell variability. We believe paired cell inpainting is a generalizable method to obtain feature representations of single cells in multichannel microscopy images.

Highlights

Feature representations of cells within microscopy images are critical for quantifying cell biology in an objective way
To understand the cell biology captured by microscopy images, researchers use features, or measurements of relevant properties of cells, such as the shape or size of cells, or the intensity of fluorescent markers
Deep learning techniques based on convolutional neural networks (CNNs) automatically learn features, which can outperform manually-defined features at image analysis tasks

Summary

Introduction

Feature representations of cells within microscopy images are critical for quantifying cell biology in an objective way. By extracting a range of different features, an image of a cell can be represented as a set of values: these feature representations can be used for numerous downstream applications, such as classifying the effects of pharmaceuticals on cancer cells [3], or exploratory analyses of protein localization [1,4]. The success of these applications depends highly on the quality of the features used: good features are challenging to define, as they must be sensitive to differences in biology, but robust to nuisance variation such as microscopy illumination effects or single cell variation [5]. The features learned by CNNs are thought to be more sensitive to relevant image content than human-designed features, offering a promising alternative for feature-based image analysis applications

Objectives

Methods

Results

Discussion

Conclusion