Machine learning in cell biology - teaching computers to recognize phenotypes

PopulationProfiler: A Tool for Population Analysis and Visualization of Image-Based Cell Screening Data

Image-based screening typically produces quantitative measurements of cell appearance. Large-scale screens involving tens of thousands of images, each containing hundreds of cells described by hundreds of measurements, result in overwhelming amounts of data. Reducing per-cell measurements to the averages across the image(s) for each treatment leads to loss of potentially valuable information on population variability. We present PopulationProfiler—a new software tool that reduces per-cell measurements to population statistics. The software imports measurements from a simple text file, visualizes population distributions in a compact and comprehensive way, and can create gates for subpopulation classes based on control samples. We validate the tool by showing how PopulationProfiler can be used to analyze the effect of drugs that disturb the cell cycle, and compare the results to those obtained with flow cytometry.

Microfluidics for High-Throughput Quantitative Studies of Early Development

Developmental biology has traditionally relied on qualitative analyses; recently, however, as in other fields of biology, researchers have become increasingly interested in acquiring quantitative knowledge about embryogenesis. Advances in fluorescence microscopy are enabling high-content imaging in live specimens. At the same time, microfluidics and automation technologies are increasing experimental throughput for studies of multicellular models of development. Furthermore, computer vision methods for processing and analyzing bioimage data are now leading the way toward quantitative biology. Here, we review advances in the areas of fluorescence microscopy, microfluidics, and data analysis that are instrumental to performing high-content, high-throughput studies in biology and specifically in development. We discuss a case study of how these techniques have allowed quantitative analysis and modeling of pattern formation in the Drosophila embryo.

Machine Learning for High-Throughput Stress Phenotyping in Plants

Advances in automated and high-throughput imaging technologies have resulted in a deluge of high-resolution images and sensor data of plants. However, extracting patterns and features from this large corpus of data requires the use of machine learning (ML) tools to enable data assimilation and feature identification for stress phenotyping. Four stages of the decision cycle in plant stress phenotyping and plant breeding activities where different ML approaches can be deployed are (i) identification, (ii) classification, (iii) quantification, and (iv) prediction (ICQP). We provide here a comprehensive overview and user-friendly taxonomy of ML tools to enable the plant community to correctly and easily apply the appropriate ML tools and best-practice guidelines for various biotic and abiotic stress traits.

Semi-automated quantification of living cells with internalized nanostructures

BACKGROUND

Nanostructures fabricated by different methods have become increasingly important for various applications in biology and medicine, such as agents for medical imaging or cancer therapy. In order to understand their interaction with living cells and their internalization kinetics, several attempts have been made in tagging them. Although methods have been developed to measure the number of nanostructures internalized by the cells, there are only few approaches aimed to measure the number of cells that internalize the nanostructures, and they are usually limited to fixed-cell studies. Flow cytometry can be used for live-cell assays on large populations of cells, however it is a single time point measurement, and does not include any information about cell morphology. To date many of the observations made on internalization events are limited to few time points and cells.

RESULTS

In this study, we present a method for quantifying cells with internalized magnetic nanowires (NWs). A machine learning-based computational framework, CellCognition, is adapted and used to classify cells with internalized and no internalized NWs, labeled with the fluorogenic pH-dependent dye pHrodo™ Red, and subsequently to determine the percentage of cells with internalized NWs at different time points. In a "proof-of-concept", we performed a study on human colon carcinoma HCT 116 cells and human epithelial cervical cancer HeLa cells interacting with iron (Fe) and nickel (Ni) NWs.

CONCLUSIONS

This study reports a novel method for the quantification of cells that internalize a specific type of nanostructures. This approach is suitable for high-throughput and real-time data analysis and has the potential to be used to study the interaction of different types of nanostructures in live-cell assays.

Microscopy Image Browser: A Platform for Segmentation and Analysis of Multidimensional Datasets

Understanding the structure-function relationship of cells and organelles in their natural context requires multidimensional imaging. As techniques for multimodal 3-D imaging have become more accessible, effective processing, visualization, and analysis of large datasets are posing a bottleneck for the workflow. Here, we present a new software package for high-performance segmentation and image processing of multidimensional datasets that improves and facilitates the full utilization and quantitative analysis of acquired data, which is freely available from a dedicated website. The open-source environment enables modification and insertion of new plug-ins to customize the program for specific needs. We provide practical examples of program features used for processing, segmentation and analysis of light and electron microscopy datasets, and detailed tutorials to enable users to rapidly and thoroughly learn how to use the program.

Quantifying co-cultured cell phenotypes in high-throughput using pixel-based classification

Biologists increasingly use co-culture systems in which two or more cell types are grown in cell culture together in order to better model cells' native microenvironments. Co-cultures are often required for cell survival or proliferation, or to maintain physiological functioning in vitro. Having two cell types co-exist in culture, however, poses several challenges, including difficulties distinguishing the two populations during analysis using automated image analysis algorithms. We previously analyzed co-cultured primary human hepatocytes and mouse fibroblasts in a high-throughput image-based chemical screen, using a combination of segmentation, measurement, and subsequent machine learning to score each cell as hepatocyte or fibroblast. While this approach was successful in counting hepatocytes for primary screening, segmentation of the fibroblast nuclei was less accurate. Here, we present an improved approach that more accurately identifies both cell types. Pixel-based machine learning (using the software ilastik) is used to seed segmentation of each cell type individually (using the software CellProfiler). This streamlined and accurate workflow can be carried out using freely available and open source software.

Cell-based screens and phenomics with fission yeast

Next-generation sequencing approaches have considerably advanced our understanding of genome function and regulation. However, the knowledge of gene function and complex cellular processes remains a challenge and bottleneck in biological research. Phenomics is a rapidly emerging area, which seeks to rigorously characterize all phenotypes associated with genes or gene variants. Such high-throughput phenotyping under different conditions can be a potent approach toward gene function. The fission yeast Schizosaccharomyces pombe (S. pombe) is a proven eukaryotic model organism that is increasingly used for genomewide screens and phenomic assays. In this review, we highlight current large-scale, cell-based approaches used with S. pombe, including computational colony-growth measurements, genetic interaction screens, parallel profiling using barcodes, microscopy-based cell profiling, metabolomic methods and transposon mutagenesis. These diverse methods are starting to offer rich insights into the relationship between genotypes and phenotypes.

Live imaging of muscles in Drosophila metamorphosis: Towards high-throughput gene identification and function analysis

Time-lapse microscopy in developmental biology is an emerging tool for functional genomics. Phenotypic effects of gene perturbations can be studied non-invasively at multiple time points in chronological order. During metamorphosis of Drosophila melanogaster, time-lapse microscopy using fluorescent reporters allows visualization of alternative fates of larval muscles, which are a model for the study of genes related to muscle wasting. While doomed muscles enter hormone-induced programmed cell death, a smaller population of persistent muscles survives to adulthood and undergoes morphological remodeling that involves atrophy in early, and hypertrophy in late pupation. We developed a method that combines in vivo imaging, targeted gene perturbation and image analysis to identify and characterize genes involved in muscle development. Macrozoom microscopy helps to screen for interesting muscle phenotypes, while confocal microscopy in multiple locations over 4-5 days produces time-lapse images that are used to quantify changes in cell morphology. Performing a similar investigation using fixed pupal tissues would be too time-consuming and therefore impractical. We describe three applications of our pipeline. First, we show how quantitative microscopy can track and measure morphological changes of muscle throughout metamorphosis and analyze genes involved in atrophy. Second, our assay can help to identify genes that either promote or prevent histolysis of abdominal muscles. Third, we apply our approach to test new fluorescent proteins as live markers for muscle development. We describe mKO2 tagged Cysteine proteinase 1 (Cp1) and Troponin-I (TnI) as examples of proteins showing developmental changes in subcellular localization. Finally, we discuss strategies to improve throughput of our pipeline to permit genome-wide screens in the future.

An Interactive Learning Framework for Scalable Classification of Pathology Images

Recent advances in microscopy imaging and genomics have created an explosion of patient data in the pathology domain. Whole-slide images (WSIs) of tissues can now capture disease processes as they unfold in high resolution, recording the visual cues that have been the basis of pathologic diagnosis for over a century. Each WSI contains billions of pixels and up to a million or more microanatomic objects whose appearances hold important prognostic information. Computational image analysis enables the mining of massive WSI datasets to extract quantitative morphologic features describing the visual qualities of patient tissues. When combined with genomic and clinical variables, this quantitative information provides scientists and clinicians with insights into disease biology and patient outcomes. To facilitate interaction with this rich resource, we have developed a web-based machine-learning framework that enables users to rapidly build classifiers using an intuitive active learning process that minimizes data labeling effort. In this paper we describe the architecture and design of this system, and demonstrate its effectiveness through quantification of glioma brain tumors.

Morphological Profiles of RNAi-Induced Gene Knockdown Are Highly Reproducible but Dominated by Seed Effects

RNA interference and morphological profiling-the measurement of thousands of phenotypes from individual cells by microscopy and image analysis-are a potentially powerful combination. We show that morphological profiles of RNAi-induced knockdown using the Cell Painting assay are in fact highly sensitive and reproducible. However, we find that the magnitude and prevalence of off-target effects via the RNAi seed-based mechanism make morphological profiles of RNAi reagents targeting the same gene look no more similar than reagents targeting different genes. Pairs of RNAi reagents that share the same seed sequence produce image-based profiles that are much more similar to each other than profiles from pairs designed to target the same gene, a phenomenon previously observed in small-scale gene-expression profiling experiments. Various strategies have been used to enrich on-target versus off-target effects in the context of RNAi screening where a narrow set of phenotypes are measured, mostly based on comparing multiple sequences targeting the same gene; however, new approaches will be needed to make RNAi morphological profiling (that is, comparing multi-dimensional phenotypes) viable. We have shared our raw data and computational pipelines to facilitate research.

Static image analysis as new approach for the characterization of tumor cell lysate used in dendritic cell vaccine preparation

BACKGROUND

Safety is an important consideration for the clinical application of dendritic cells (DC) loaded with autologous tumor lysate (TL). Thus, avitalization of TL from living autologous tumor tissue has to be guaranteed.

METHODS

Composition of TL was investigated by static image analysis (SIA) with the Morphologi G3 device, which simultaneously measures size and shape of up to 100,000 particles within one sample run. This approach was compared with sample characterization by high-resolution automated cell counting, trypan blue staining, and ATP quantification.

RESULTS

Using SIA, we only detected fragmented, non-cellular structures in completely avitalized TL, indicating complete destruction of living cells. Analysis of particle size distribution by SIA as well as CASY cell counter showed that 95% of particles had a diameter of <10 µm as a sign of cell fragmentation. Complete avitalization of TL was confirmed with trypan blue staining and ATP analysis.

CONCLUSION

Regarding generation of DC vaccines, the proof of avitality of TL from living tumor tissue can clearly be achieved by SIA alone or in combination with standard assays. Our data show that SIA is a highly precise method for TL characterization. The SIA device complies with FDA regulation and, therefore, might be suitable for characterization of cellular therapy medicinal products.

High-throughput fluorescence correlation spectroscopy enables analysis of proteome dynamics in living cells

To understand the function of cellular protein networks, spatial and temporal context is essential. Fluorescence correlation spectroscopy (FCS) is a single-molecule method to study the abundance, mobility and interactions of fluorescence-labeled biomolecules in living cells. However, manual acquisition and analysis procedures have restricted live-cell FCS to short-term experiments of a few proteins. Here, we present high-throughput (HT)-FCS, which automates screening and time-lapse acquisition of FCS data at specific subcellular locations and subsequent data analysis. We demonstrate its utility by studying the dynamics of 53 nuclear proteins. We made 60,000 measurements in 10,000 living human cells, to obtain biophysical parameters that allowed us to classify proteins according to their chromatin binding and complex formation. We also analyzed the cell-cycle-dependent dynamics of the mitotic kinase complex Aurora B/INCENP and showed how a rise in Aurora concentration triggers two-step complex formation. We expect that throughput and robustness will make HT-FCS a broadly applicable technology for characterizing protein network dynamics in cells.

Toward high-content screening of mitochondrial morphology and membrane potential in living cells

Mitochondria are double membrane organelles involved in various key cellular processes. Governed by dedicated protein machinery, mitochondria move and continuously fuse and divide. These "mitochondrial dynamics" are bi-directionally linked to mitochondrial and cell functional state in space and time. Due to the action of the electron transport chain (ETC), the mitochondrial inner membrane displays a inside-negative membrane potential (Δψ). The latter is considered a functional readout of mitochondrial "health" and required to sustain normal mitochondrial ATP production and mitochondrial fusion. During the last decade, live-cell microscopy strategies were developed for simultaneous quantification of Δψ and mitochondrial morphology. This revealed that ETC dysfunction, changes in Δψ and aberrations in mitochondrial structure often occur in parallel, suggesting they are linked potential targets for therapeutic intervention. Here we discuss how combining high-content and high-throughput strategies can be used for analysis of genetic and/or drug-induced effects at the level of individual organelles, cells and cell populations. This article is part of a Directed Issue entitled: Energy Metabolism Disorders and Therapies.

Quantifying small molecule phenotypic effects using mitochondrial morpho-functional fingerprinting and machine learning

In primary fibroblasts from Leigh Syndrome (LS) patients, isolated mitochondrial complex I deficiency is associated with increased reactive oxygen species levels and mitochondrial morpho-functional changes. Empirical evidence suggests these aberrations constitute linked therapeutic targets for small chemical molecules. However, the latter generally induce multiple subtle effects, meaning that in vitro potency analysis or single-parameter high-throughput cell screening are of limited use to identify these molecules. We combine automated image quantification and artificial intelligence to discriminate between primary fibroblasts of a healthy individual and a LS patient based upon their mitochondrial morpho-functional phenotype. We then evaluate the effects of newly developed Trolox variants in LS patient cells. This revealed that Trolox ornithylamide hydrochloride best counterbalanced mitochondrial morpho-functional aberrations, effectively scavenged ROS and increased the maximal activity of mitochondrial complexes I, IV and citrate synthase. Our results suggest that Trolox-derived antioxidants are promising candidates in therapy development for human mitochondrial disorders.

Semi-automatic organelle detection on transmission electron microscopic images

Recent advances in the acquisition of large-scale datasets of transmission electron microscope images have allowed researchers to determine the number and the distribution of subcellular ultrastructures at both the cellular level and the tissue level. For this purpose, it would be very useful to have a computer-assisted system to detect the structures of interest, such as organelles. Using our original image recognition framework CARTA (Clustering-Aided Rapid Training Agent), combined with procedures to highlight and enlarge regions of interest on the image, we have developed a successful method for the semi-automatic detection of plant organelles including mitochondria, amyloplasts, chloroplasts, etioplasts, and Golgi stacks in transmission electron microscope images. Our proposed semi-automatic detection system will be helpful for labelling organelles in the interpretation and/or quantitative analysis of large-scale electron microscope imaging data.

A genomic Multiprocess survey of machineries that control and link cell shape, microtubule organization, and cell-cycle progression

Understanding cells as integrated systems requires that we systematically decipher how single genes affect multiple biological processes and how processes are functionally linked. Here, we used multiprocess phenotypic profiling, combining high-resolution 3D confocal microscopy and multiparametric image analysis, to simultaneously survey the fission yeast genome with respect to three key cellular processes: cell shape, microtubule organization, and cell-cycle progression. We identify, validate, and functionally annotate 262 genes controlling specific aspects of those processes. Of these, 62% had not been linked to these processes before and 35% are implicated in multiple processes. Importantly, we identify a conserved role for DNA-damage responses in controlling microtubule stability. In addition, we investigate how the processes are functionally linked. We show unexpectedly that disruption of cell-cycle progression does not necessarily affect cell size control and that distinct aspects of cell shape regulate microtubules and vice versa, identifying important systems-level links across these processes.

FindFoci: a focus detection algorithm with automated parameter training that closely matches human assignments, reduces human inconsistencies and increases speed of analysis

Accurate and reproducible quantification of the accumulation of proteins into foci in cells is essential for data interpretation and for biological inferences. To improve reproducibility, much emphasis has been placed on the preparation of samples, but less attention has been given to reporting and standardizing the quantification of foci. The current standard to quantitate foci in open-source software is to manually determine a range of parameters based on the outcome of one or a few representative images and then apply the parameter combination to the analysis of a larger dataset. Here, we demonstrate the power and utility of using machine learning to train a new algorithm (FindFoci) to determine optimal parameters. FindFoci closely matches human assignments and allows rapid automated exploration of parameter space. Thus, individuals can train the algorithm to mirror their own assignments and then automate focus counting using the same parameters across a large number of images. Using the training algorithm to match human assignments of foci, we demonstrate that applying an optimal parameter combination from a single image is not broadly applicable to analysis of other images scored by the same experimenter or by other experimenters. Our analysis thus reveals wide variation in human assignment of foci and their quantification. To overcome this, we developed training on multiple images, which reduces the inconsistency of using a single or a few images to set parameters for focus detection. FindFoci is provided as an open-source plugin for ImageJ.

Single nucleotide variations: biological impact and theoretical interpretation

Genome-wide association studies (GWAS) and whole-exome sequencing (WES) generate massive amounts of genomic variant information, and a major challenge is to identify which variations drive disease or contribute to phenotypic traits. Because the majority of known disease-causing mutations are exonic non-synonymous single nucleotide variations (nsSNVs), most studies focus on whether these nsSNVs affect protein function. Computational studies show that the impact of nsSNVs on protein function reflects sequence homology and structural information and predict the impact through statistical methods, machine learning techniques, or models of protein evolution. Here, we review impact prediction methods and discuss their underlying principles, their advantages and limitations, and how they compare to and complement one another. Finally, we present current applications and future directions for these methods in biological research and medical genetics.