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

Abstract

Deep profiling of cell states can provide a broad picture of biological changes that occur in disease, mutation, or in response to drug or chemical treatments. Morphological and gene expression profiling, for example, can cost-effectively capture thousands of features in thousands of samples across perturbations, but it is unclear to what extent the two modalities capture overlapping versus complementary mechanistic information. Here, using both the L1000 and Cell Painting assays to profile gene expression and cell morphology, respectively, we perturb A549 lung cancer cells with 1,327 small molecules from the Drug Repurposing Hub across six doses. We determine that the two assays capture some shared and some complementary information in mapping cell state. We find that as compared to L1000, Cell Painting captures a higher proportion of reproducible compounds and has more diverse samples, but measures fewer distinct groups of features. In an unsupervised analysis, Cell Painting grouped more compound mechanisms of action (MOA) whereas in a supervised deep learning analysis, L1000 predicted more MOAs. In general, the two assays together provide a complementary view of drug mechanisms for follow up analyses. Our analyses answer fundamental biological questions comparing the two biological modalities and, given the numerous applications of profiling in biology, provide guidance for planning experiments that profile cells for detecting distinct cell types, disease phenotypes, and response to chemical or genetic perturbations.

Predicting drug polypharmacology from cell morphology readouts using variational autoencoder latent space arithmetic

“…High test statistics for L2 distance and a low test statistics for Pearson correlation indicates that the specific MOA “A ∩ B” could not be predicted, either because of incorrect annotations, non-additive or synergistic treatment effects, or a low penetrant phenotype unable to be captured in Cell Painting data. Compared to L1000 gene expression profiles, we observed that gene expression and morphology assays are complementary, but also predict many of the same polypharmacology MOAs ( Fig 5 ) , which is consistent with recent work [ 25 ].…”

“…Because we had access to the same perturbations with L1000 readouts, we were able to compare cell morphology and gene expression results. We found that both models capture complementary information when predicting polypharmacology, which is a similar observation to a deeper dive which compared the different technology information content [25]. We did not explore multi-modal data integration in this project, as this has been explored in more detail in other recent publications [28,29].…”

Cell Painting predicts impact of lung cancer variants

“…Because the profiles of most variants tend to cluster together within each gene, as observed in the hierarchical clustering of the correlation matrix (Figure 1F), we conclude that the phenotypic variations of alleles remain closely related to the reference gene and rarely result in a major phenotypic disruption that places them in a different cluster. This type of closely related variation is consistent with previous studies in morphological and transcriptional profiling 14,17 , which report that the major factor of variation detected by profiling platforms is first associated with cell lines, then with groups of perturbations that share similar mechanisms, and finally with specific effects of each perturbation.…”

“…We evaluated this as follows: after acquiring Cell Painting images for each sample (Figure 1B), we transformed them into replicate-level allele profiles using a deep learning-based workflow 12,13 (Figure 1A, see also Methods). We evaluated the quality of profiles using the percent replicating score 14 , measured as the percentage of perturbations whose replicates consistently have higher similarity (reproducible signal) than random sets of perturbations; in this case 83.2% (Figure 1C). 1080x1080), and each channel has been independently rescaled to fit the visible intensity range.…”

“…As is the case in many biological experiments, imaging assays may also be prone to nuisance variation due to technical artifacts. We used negative control spherizing to correct for batch effect biases, which has shown to be effective in other studies 13,14,32 . The spherizing transform used in this work makes the assumption that negative controls sampled from different batches ought to be similar to each other in the biological sense, and any deviations from this normal looking phenotype is rather technical.…”