Explore a 42-minute lecture from the Broad Institute's Primer on Medical and Population Genetics series, focusing on learning representations for image-based profiling of perturbations. Delve into topics such as the Cell Painting assay, image-based morphology profiling, predicting assay-compound interactions and cell health phenotypes, and lung cancer variant impact prediction. Examine various sources of features, including classical features for profiling cell state and transfer learning with pre-trained CNNs. Investigate weakly supervised learning (WSL) and its confounders, single-cell classification results, and batch-correction techniques. Learn about perturbation selection, combined Cell Painting datasets, and the training and performance of Cell Painting CNN-1. Analyze the effects of batch correction and gain insights into the qualitative analysis and availability of Cell Painting CNN-1. This comprehensive lecture, presented by Nikita Moshkov from the Caicedo Lab, offers valuable knowledge for researchers, technicians, students, and investigators in the field of genetics and image-based profiling.
Learning Representations for Image-Based Profiling of Perturbations - MPG Primer 2023
Overview
Syllabus
Intro
Cell Painting assay
Image-based morphology profiling: introduction
Predicting assay-compound interaction
Predicting cell health phenotypes
Lung cancer variant impact prediction
Sources of features
Profiling cell state: classical features
Transfer learning: pre-trained CNNs
Weakly supervised learning (WSL)
Confounders for WSL
Single-cell classification results
Batch-correction with sphering
Perturbation selection
Combined Cell Painting dataset
Cell Painting CNN-1 training
Cell Painting CNN-1 performance
Effect of batch correction
Cell Painting CNN-1 qualitative analysis
Cell Painting CNN-1 availability
Taught by
Broad Institute
