Predicting gene expression from cell morphology in human induced pluripotent stem cells

Abstract

Purification is essential before differentiating human induced pluripotent stem cells (hiPSCs) into cells that fully express particular differentiation marker genes. High-quality iPSC clones are typically purified through gene expression profiling or visual inspection of the cell morphology; however, the relationship between the two methods remains unclear. We investigated the relationship between gene expression levels and morphology by analyzing live-cell phase-contrast images and mRNA profiles collected during the purification process. We employed this data and an unsupervised image feature extraction method to build a model that predicts gene expression levels from morphology. As a benchmark, we confirmed that the method can predict the gene expression levels from tissue images for cancer genes, performing as well as state-of-the-art methods. We then applied the method to iPSCs and identified two genes that are well-predicted from cell morphology. Although strong batch effects resulting from the reprogramming process preclude the ability to use the same model to predict across batches, prediction within a reprogramming batch is sufficiently robust to provide a practical approach for estimating expression levels of a few genes and monitoring the purification process.

Morphology and gene expression profiling provide complementary information for mapping cell state

“…Integrating gene expression and morphology profiles with chemical structures revealed that each data type provides complementary information for predicting a drug’s mechanism of action (Haghighi et al, 2021; Nassiri and McCall, 2018), for predicting the effects of perturbations (Caicedo et al, 2021a), and for identifying nuisance compounds that can lead to false hits (Dahlin et al, 2021). As well, to some degree, gene expression and morphology datasets contain sufficient information to predict changes in each other (Haghighi et al, 2021; Nassiri and McCall, 2018; Wakui et al, 2022).…”

Deep learning identifies heterogeneous subpopulations in breast cancer cell lines

“…The ability to autonomously observe shifts and correlate them to specific events could enable researchers to observe cell health, view the effects of different drug mechanisms, and observe phenotypic plasticity without the need for techniques such as sequencing or fluorescent labeling. One method that has been used on cell populations is autoencoders 16,21,22,44 , which contain a reduced latent space that is representative of the input and have been used for unsupervised learning approaches. Autoencoders have been used primarily to perform latent space predictions and provide explanations for changes in cell morphology.…”

“…Recently, image analysis using machine learning has demonstrated that cell-state properties such as metastatic invasion [16][17][18] , the induction of an epithelial to mesenchymal transition 19 , or the introduction of genetic perturbations 20 can be detected through cellular morphology. Notably, deep learning has been effective at learning relevant biological features [21][22][23][24] . Many of these results have been demonstrated only using basic imaging techniques, such as phase contrast or brightfield microscopy.…”

Uncovering features of synapses in primary visual cortex through contrastive representation learning

“…Specific targets for either input labels or annotations could include electrical synapses 74 or more subtle synaptic organelles like multivesicular-endoplasmic reticulum bodies 75 . Given that transcriptional states are strongly correlated with morphological features in cells and tissues [76][77][78][79] , we believe that informative morphological representations such as those produced with SynapseCLR may help form a more complete picture about the connection between structural connectivity and the underlying molecular mechanisms.…”