Learning representations for image-based profiling of perturbations

Abstract

Measuring the phenotypic effect of treatments on cells through imaging assays is an efficient and powerful way of studying cell biology, and requires computational methods for transforming images into quantitative data that highlights phenotypic outcomes. Here, we present an optimized strategy for learning representations of treatment effects from high-throughput imaging data, which follows a causal framework for interpreting results and guiding performance improvements. We use weakly supervised learning (WSL) for modeling associations between images and treatments, and show that it encodes both confounding factors and phenotypic features in the learned representation. To facilitate their separation, we constructed a large training dataset with Cell Painting images from five different sources to maximize experimental diversity, following insights from our causal analysis. Training a WSL model with this dataset successfully improves downstream performance, and produces a reusable convolutional network for image-based profiling, which we call Cell Painting CNN. We conducted a comprehensive evaluation of our strategy on three publicly available Cell Painting datasets, discovering that representations obtained by the Cell Painting CNN can improve performance in downstream analysis up to 25% with respect to classical features, while also being more computationally efficient.

Deep Representation Learning Determines Drug Mechanism of Action from Cell Painting Images

“…Accurate performance on test wells never seen during the model’s training suggests that the model was not leveraging well-specific features for its learning, which can be a confounder in microscopy 34 . Furthermore, we ensured that each MOA’s test set wells were spread across multiple plates (Figure 2D).…”

“…Here, we present a MOA determination method spanning hundreds of MOAs that showed efficacy on two independent datasets: 1) the Joint Undertaking in Morphological Profiling (JUMP1) pilot dataset 30 encompassing 176 MOAs 2) the Library of Integrated Network-Based Cellular Signatures (LINCS) dataset 31,32 encompassing 424 MOAs. We compared our method called MOAProfiler (MP) with CP as well as a deep learning based method called DeepProfiler (DP) 33,34 . We present MP as an open-source and readily available tool for deep phenotypic profiling of Cell Painting images.…”

RxRx3: Phenomics Map of Biology

“…Progress toward general purpose deep learning frameworks for HCS have been driven by weakly-supervised learning and more recently by self-supervised learning (SSL). In weakly-supervised learning 9,21 , perturbation metadata (i.e. the perturbation applied to each well) is used to label HCS images and train supervised deep learning models.…”

RxRx1: A Dataset for Evaluating Experimental Batch Correction Methods

“…Our approach relies on weakly supervised learning [47,48] since we train models to predict the experimental perturbation in each well, without validating that each treatment induces a unique visual phenotype (N.B. : such validation is likely impossible).…”