Applications in image-based profiling of perturbations

1. Download : Download high-res image (191KB)
2. Download : Download full-size image

     

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

A variational autoencoder (VAE) is a machine learning algorithm, useful for generating a compressed and interpretable latent space. These representations have been generated from various biomedical data types and can be used to produce realistic-looking simulated data. However, standard vanilla VAEs suffer from entangled and uninformative latent spaces, which can be mitigated using other types of VAEs such as β-VAE and MMD-VAE. In this project, we evaluated the ability of VAEs to learn cell morphology characteristics derived from cell images. We trained and evaluated these three VAE variants—Vanilla VAE, β-VAE, and MMD-VAE—on cell morphology readouts and explored the generative capacity of each model to predict compound polypharmacology (the interactions of a drug with more than one target) using an approach called latent space arithmetic (LSA). To test the generalizability of the strategy, we also trained these VAEs using gene expression data of the same compound perturbations and found that gene expression provides complementary information. We found that the β-VAE and MMD-VAE disentangle morphology signals and reveal a more interpretable latent space. We reliably simulated morphology and gene expression readouts from certain compounds thereby predicting cell states perturbed with compounds of known polypharmacology. Inferring cell state for specific drug mechanisms could aid researchers in developing and identifying targeted therapeutics and categorizing off-target effects in the future.     

Morphodynamic profiling of protrusion phenotypes

We propose a framework for tracking arbitrary complex cell boundary movements, relying on a unique definition of protrusion and retraction as the pathlength a virtual edge marker traverses when moving continuously perpendicular to the cell boundary. We introduce the level set method as a numerical scheme to reconstruct continuous boundary movement in time-lapse image sequences with finite time sampling. For moderately complex movements, we describe a numerically less expensive method that satisfactorily approximates the definition. Densely sampled protrusion and retraction rates were accumulated in space-time charts revealing distinct morphodynamic states. Applying this technique to the profiling of epithelial cell protrusion we identified three different states. In the I-state, long cell edge sectors are synchronized in cycles of protrusion and retraction. In the V-state random bursts of protrusion initiate protrusion waves propagating transversally in both directions. Cells switch between both states dependent on the Rac1 activation level. Furthermore, the persistence of transversal waves in the V-state depends on Arp2/3 concentration. Inhibition of PAK shifts cells into a lambda-state where continuous protrusion is occasionally interrupted by self-propagating ruffles. Our data support a model where activation of Rac1 mediates the propagation of protrusion waves, whose persistence depends on the relative abundance of activated Arp2/3 and polymerizable G-actin.     

Microglial cell origin and phenotypes in health and disease

Microglia - resident myeloid-lineage cells in the brain and the spinal cord parenchyma - function in the maintenance of normal tissue homeostasis. Microglia also act as sentinels of infection and injury, and participate in both innate and adaptive immune responses in the central nervous system. Microglia can become activated and/or dysregulated in the context of neurodegenerative disease and cancer, and thereby contribute to disease severity. Here, we discuss recent studies that provide new insights into the origin and phenotypes of microglia in health and disease.     

Systematic profiling of cellular phenotypes with spotted cell microarrays reveals mating-pheromone response genes

We have developed spotted cell microarrays for measuring cellular phenotypes on a large scale. Collections of cells are printed, stained for subcellular features, then imaged via automated, high-throughput microscopy, allowing systematic phenotypic characterization. We used this technology to identify genes involved in the response of yeast to mating pheromone. Besides morphology assays, cell microarrays should be valuable for high-throughput in situ hybridization and immunoassays, enabling new classes of genetic assays based on cell imaging.     

Selection of carbohydrate-binding cell phenotypes using oligosaccharide-coated magnetic particles

Neoglycoconjugate coated magnetic beads were assessed for theirability to selectively isolate human cells with known anti-carbohydratereactivity. Four lung cancer cell lines, NCI-H146, NCI-N417D,SKMES-1, EKVX; two acute lymphoblastic leukemia lines, MOLT-4and CCRF-CEM; and the anti- Le^c (isolactosamine) hybridoma,LU-BCRU-G7, were tested. The neoglycoconjugates (biotinylatedpseudopolysaccharides) bound uniformly to streptavidin coatedmagnetic beads as demonstrated by FTTC labeled lectin. Streptavidinbeads alone did not bind to any of the cell types. The anti-Le^c hybridoma cell line, LU BCRU-G7, demonstrated binding onlyto Le^c pseudopolysaccharide coated magnetic beads. Subsequentincubation in the presence of unlabeled pseudopolysaccharideresulted in the release of the beads from the cell surface.Although there was some heterogeneity within the individuallung and leukemic cell lines, positive cells showed strong rosetteformation with the coated beads. The A_dl disaccharide coatedbeads showed binding in all four lung cancer cell lines, withthe Le^c and the H (type1) pseudopolysaccharide-bead conjugatesonly reactive in the N417 and H146 SCLC lines. The range ofL-selectin ligand-coated beads were all successful in bindingto the acute lymphoblastic leukemia cell lines MOLT4 and CCRF-CEM.This approach provides a versatile model for the study of cell-surfacecarbohydrate interactions that should find application in manyareas of cell biology. carbohydrate-coated magnetic beads cell selec-tion lectins neoglycoconjugates pseudopolysaccharides     

Quality prediction of cell substrate using gene expression profiling

Changes in cell culture conditions influence the metabolism of cells, which consequently affects the quality of the products that they produce, such as viral vectors, recombinant proteins, or vaccines. Currently there is no effective technique available to monitor global quality of cells in cell culture. Here we describe a new method using gene expression profiling by microarray to predict the quality of cell substrates. Human embryonic kidney 293 cells are a commonly used cell substrate in the production of biological products. We demonstrate that the yield of adenoviral vectors was lower in over-confluent 293 cells, compared to 40 or 90% confluent cells. Total RNA derived from these cells of different confluence states was reverse transcribed, labeled, and used to hybridize 10K cDNA arrays to determine biomarkers for confluence states. Phenotype scatter-plot analysis and cluster analysis were used for class discovery. Based on this approach, we identified genes that were either up-regulated or down-modulated in response to different cell confluence states. By multivariate predictive models we identified a set of 37 genes that were either down-regulated or up-regulated compared to 90% confluent cells as a predictor of cell confluence and quality of 293 cell cultures. The predictive accuracy of these models was assessed by the leave-one-out cross-validation method. The expression of selected gene predictors was validated by quantitative PCR analysis. Our results demonstrate that gene expression profiling can assess the quality of cell substrates prior to large-scale production of a biological product.     

Predicting protein phenotypes based on protein-protein interaction network

Background

Identifying associated phenotypes of proteins is a challenge of the modern genetics since the multifactorial trait often results from contributions of many proteins. Besides the high-through phenotype assays, the computational methods are alternative ways to identify the phenotypes of proteins.

Methodology/Principal Findings

Here, we proposed a new method for predicting protein phenotypes in yeast based on protein-protein interaction network. Instead of only the most likely phenotype, a series of possible phenotypes for the query protein were generated and ranked acording to the tethering potential score. As a result, the first order prediction accuracy of our method achieved 65.4% evaluated by Jackknife test of 1,267 proteins in budding yeast, much higher than the success rate (15.4%) of a random guess. And the likelihood of the first 3 predicted phenotypes including all the real phenotypes of the proteins was 70.6%.

Conclusions/Significance

The candidate phenotypes predicted by our method provided useful clues for the further validation. In addition, the method can be easily applied to the prediction of protein associated phenotypes in other organisms.     

Virtual profiling: a new way to analyse phenotypes

Simulation models can be used to perform virtual profiling in order to analyse eco‐physiological processes controlling plant phenotype. To illustrate this, an eco‐physiological model has been used to compare and contrast the status of a virtual fruit system under two situations of carbon supply. The model simulates fruit growth, accumulation of sugar, citric acid and water, transpiration, respiration and ethylene emission, and was successfully tested on peach (Prunus persica L. Batsch) for two leaf‐to‐fruit ratios (6 and 18 leaves per fruit). The development stage and the variation in leaf number had large effects of the fruit model variables dealing with growth, metabolism and fruit quality. A sensitivity analysis showed that changing a single parameter value, which could correspond to a genotypic change induced by a mutation, either strongly affects most of the processes, or affects a specific process or none. Correlation analysis showed that, in a complex system such as fruit, the intensity of many physiological processes and quality traits co‐varies. It also showed unexpected co‐variations resulting from emergent properties of the system. This virtual profiling approach opens a new route to explore the impact of mutations, or naturally occurring genetic variations, under differing environmental conditions.     

DeepLearnMOR: a deep-learning framework for fluorescence image-based classification of organelle morphology

The proper biogenesis, morphogenesis, and dynamics of subcellular organelles are essential to their metabolic functions. Conventional techniques for identifying, classifying, and quantifying abnormalities in organelle morphology are largely manual and time-consuming, and require specific expertise. Deep learning has the potential to revolutionize image-based screens by greatly improving their scope, speed, and efficiency. Here, we used transfer learning and a convolutional neural network (CNN) to analyze over 47,000 confocal microscopy images from Arabidopsis wild-type and mutant plants with abnormal division of one of three essential energy organelles: chloroplasts, mitochondria, or peroxisomes. We have built a deep-learning framework, DeepLearnMOR (Deep Learning of the Morphology of Organelles), which can rapidly classify image categories and identify abnormalities in organelle morphology with over 97% accuracy. Feature visualization analysis identified important features used by the CNN to predict morphological abnormalities, and visual clues helped to better understand the decision-making process, thereby validating the reliability and interpretability of the neural network. This framework establishes a foundation for future larger-scale research with broader scopes and greater data set diversity and heterogeneity.

An automated and explainable deep-learning framework allows rapidly classifying abnormalities in organelle morphology.     

Predicting hepatitis B virus-positive metastatic hepatocellular carcinomas using gene expression profiling and supervised machine learning

Hepatocellular carcinoma (HCC) is one of the most common and aggressive human malignancies. Its high mortality rate is mainly a result of intra-hepatic metastases. We analyzed the expression profiles of HCC samples without or with intra-hepatic metastases. Using a supervised machine-learning algorithm, we generated for the first time a molecular signature that can classify metastatic HCC patients and identified genes that were relevant to metastasis and patient survival. We found that the gene expression signature of primary HCCs with accompanying metastasis was very similar to that of their corresponding metastases, implying that genes favoring metastasis progression were initiated in the primary tumors. Osteopontin, which was identified as a lead gene in the signature, was over-expressed in metastatic HCC; an osteopontin-specific antibody effectively blocked HCC cell invasion in vitro and inhibited pulmonary metastasis of HCC cells in nude mice. Thus, osteopontin acts as both a diagnostic marker and a potential therapeutic target for metastatic HCC.     

Large-scale protein profiling in human cell lines using antibody-based proteomics

Human cancer cell lines grown in vitro are frequently used to decipher basic cell biological phenomena and to also specifically study different forms of cancer. Here we present the first large-scale study of protein expression patterns in cell lines using an antibody-based proteomics approach. We analyzed the expression pattern of 5436 proteins in 45 different cell lines using hierarchical clustering, principal component analysis, and two-group comparisons for the identification of differentially expressed proteins. Our results show that immunohistochemically determined protein profiles can categorize cell lines into groups that overall reflect the tumor tissue of origin and that hematological cell lines appear to retain their protein profiles to a higher degree than cell lines established from solid tumors. The two-group comparisons reveal well-characterized proteins as well as previously unstudied proteins that could be of potential interest for further investigations. Moreover, multiple myeloma cells and cells of myeloid origin were found to share a protein profile, relative to the protein profile of lymphoid leukemia and lymphoma cells, possibly reflecting their common dependency of bone marrow microenvironment. This work also provides an extensive list of antibodies, for which high-resolution images as well as validation data are available on the Human Protein Atlas ( www.proteinatlas.org ), that are of potential use in cell line studies.     

Characterization of rainbow trout cell lines using microsatellite DNA profiling

Ten microsatellite loci (Omy27DU,Omy325(A3)UoG, OmyFGT5TUF,OmyFGT14TUF, OmyFGT15TUF,OmyFGT23TUF, Omy77DU,Ssa20.19NUIG, Ots1BML, andOne18ASC) were amplified using the polymerase chain reaction to create genetic profiles for nine cell lines (RTG-2, RTH-149,RTL-W1,RTgill-W1, RTS-11, RTS-34st, RTP-2, RTP-91E and RTP-91F) from rainbow trout(Oncorhynchus mykiss) and one cell line (CHSE-214) from Chinook salmon (O. tschawytscha). A cell line (PHL) from anon-salmonid, the Pacific herring (Clupea harengus pallasi), was included as a control. The ten loci clearly revealed the uniqueness of each cell line, except for two cell lines (RTP-91E andRTP-91F) from the same fish. RTP-91E and RTP-91F were identical at all loci except Ssa20.19NUIG. The most useful locus for demonstrating uniqueness was Ots1BML. The information was used to demonstrate that an uncharacterized rainbow trout cell line (Clone 1A)was in fact CHSE-214, illustrating the usefulness of multiplexed microsatellites for the creation of genetic profiles for salmonid cell lines and for the testing of cell line cross-contamination. This revised version was published online in August 2006 with corrections to the Cover Date.     

Predicting phenotypes of asthma and eczema with machine learning

Background

There is increasing recognition that asthma and eczema are heterogeneous diseases. We investigated the predictive ability of a spectrum of machine learning methods to disambiguate clinical sub-groups of asthma, wheeze and eczema, using a large heterogeneous set of attributes in an unselected population. The aim was to identify to what extent such heterogeneous information can be combined to reveal specific clinical manifestations.

Methods

The study population comprised a cross-sectional sample of adults, and included representatives of the general population enriched by subjects with asthma. Linear and non-linear machine learning methods, from logistic regression to random forests, were fit on a large attribute set including demographic, clinical and laboratory features, genetic profiles and environmental exposures. Outcome of interest were asthma, wheeze and eczema encoded by different operational definitions. Model validation was performed via bootstrapping.

Results

The study population included 554 adults, 42% male, 38% previous or current smokers. Proportion of asthma, wheeze, and eczema diagnoses was 16.7%, 12.3%, and 21.7%, respectively. Models were fit on 223 non-genetic variables plus 215 single nucleotide polymorphisms. In general, non-linear models achieved higher sensitivity and specificity than other methods, especially for asthma and wheeze, less for eczema, with areas under receiver operating characteristic curve of 84%, 76% and 64%, respectively. Our findings confirm that allergen sensitisation and lung function characterise asthma better in combination than separately. The predictive ability of genetic markers alone is limited. For eczema, new predictors such as bio-impedance were discovered.

Conclusions

More usefully-complex modelling is the key to a better understanding of disease mechanisms and personalised healthcare: further advances are likely with the incorporation of more factors/attributes and longitudinal measures.

     

SCYN: single cell CNV profiling method using dynamic programming

Background

All diseases containing genetic material undergo genetic evolution and give rise to heterogeneity including cancer and infection. Although these illnesses are biologically very different, the ability for phylogenetic retrodiction based on the genomic reads is common between them and thus tree-based principles and assumptions are shared. Just as the different frequencies of tumor genomic variants presupposes the existence of multiple tumor clones and provides a handle to computationally infer them, we postulate that the different variant frequencies in viral reads offers the means to infer multiple co-infecting sublineages.

Results

We present a common methodological framework to infer the phylogenomics from genomic data, be it reads of SARS-CoV-2 of multiple COVID-19 patients or bulk DNAseq of the tumor of a cancer patient. We describe the Concerti computational framework for inferring phylogenies in each of the two scenarios.To demonstrate the accuracy of the method, we reproduce some known results in both scenarios. We also make some additional discoveries.

Conclusions

Concerti successfully extracts and integrates information from multi-point samples, enabling the discovery of clinically plausible phylogenetic trees that capture the heterogeneity known to exist both spatially and temporally. These models can have direct therapeutic implications by highlighting “birth” of clones that may harbor resistance mechanisms to treatment, “death” of subclones with drug targets, and acquisition of functionally pertinent mutations in clones that may have seemed clinically irrelevant. Specifically in this paper we uncover new potential parallel mutations in the evolution of the SARS-CoV-2 virus. In the context of cancer, we identify new clones harboring resistant mutations to therapy.

     

A high-throughput cell migration assay using scratch wound healing, a comparison of image-based readout methods

Background  

Cell migration is a complex phenomenon that requires the coordination of numerous cellular processes. Investigation of cell migration and its underlying biology is of interest to basic scientists and those in search of therapeutics. Current migration assays for screening small molecules, siRNAs, or other perturbations are difficult to perform in parallel at the scale required to screen large libraries.     

Predicting enzyme targets for cancer drugs by profiling human Metabolic reactions in NCI-60 cell lines

Background  

Drugs can influence the whole metabolic system by targeting enzymes which catalyze metabolic reactions. The existence of interactions between drugs and metabolic reactions suggests a potential way to discover drug targets.     

Modulation of trophoblast stem cell and giant cell phenotypes: analyses using the Rcho-1 cell model

Trophoblast giant cells are located at the maternal-embryonic interface and have fundamental roles in the invasive and endocrine phenotypes of the rodent placenta. In this report, we describe the experimental modulation of trophoblast stem cell and trophoblast giant cell phenotypes using the Rcho-1 trophoblast cell model. Rcho-1 trophoblast cells can be manipulated to proliferate or differentiate into trophoblast giant cells. Differentiated Rcho-1 trophoblast cells are invasive and possess an endocrine phenotype, including the production of members of the prolactin (PRL) family. Dimethyl sulfoxide (DMSO), a known differentiation-inducing agent, was found to possess profound effects on the in vitro development of trophoblast cells. Exposure to DMSO, at non-toxic concentrations, inhibited trophoblast giant cell differentiation in a dose-dependent manner. These concentrations of DMSO did not significantly affect trophoblast cell proliferation or survival. Trophoblast cells exposed to DMSO exhibited an altered morphology; they were clustered in tightly packed colonies. Trophoblast giant cell formation was disrupted, as was the expression of members of the PRL gene family. The effects of DMSO were reversible. Removal of DMSO resulted in the formation of trophoblast giant cells and expression of the PRL gene family. The phenotype of the DMSO-treated cells was further determined by examining the expression of a battery of genes characteristic of trophoblast stem cells and differentiated trophoblast cell lineages. DMSO treatment had a striking stimulatory effect on eomesodermin expression and a reciprocal inhibitory effect on Hand1 expression. In summary, DMSO reversibly inhibits trophoblast differentiation and induces a quiescent state, which mimics some but not all aspects of the trophoblast stem cell phenotype.