Combining Supervised and Unsupervised Machine Learning Methods for Phenotypic Functional Genomics Screening.

There has been an increase in the use of machine learning and artificial intelligence (AI) for the analysis of image-based cellular screens. The accuracy of these analyses, however, is greatly dependent on the quality of the training sets used for building the machine learning models. We propose that unsupervised exploratory methods should first be applied to the data set to gain a better insight into the quality of the data. This improves the selection and labeling of data for creating training sets before the application of machine learning. We demonstrate this using a high-content genome-wide small interfering RNA screen. We perform an unsupervised exploratory data analysis to facilitate the identification of four robust phenotypes, which we subsequently use as a training set for building a high-quality random forest machine learning model to differentiate four phenotypes with an accuracy of 91.1% and a kappa of 0.85. Our approach enhanced our ability to extract new knowledge from the screen when compared with the use of unsupervised methods alone.

Improving Comprehension Efficiency of High Content Screening Data Through Interactive Visualizations.

In this study, an experiment is conducted to measure the performance in speed and accuracy of interactive visualizations. A platform for interactive data visualizations was implemented using Django, D3, and Angular. Using this platform, a questionnaire was designed to measure a difference in performance between interactive and noninteractive data visualizations. In this questionnaire consisting of 12 questions, participants were given tasks in which they had to identify trends or patterns. Other tasks were directed at comparing and selecting algorithms with a certain outcome based on visualizations. All tasks were performed on high content screening data sets with the help of visualizations. The difference in time to carry out tasks and accuracy of performance was measured between a group viewing interactive visualizations and a group viewing noninteractive visualizations. The study shows a significant advantage in time and accuracy in the group that used interactive visualizations over the group that used noninteractive visualizations. In tasks comparing results of different algorithms, a significant decrease in time was observed in using interactive visualizations over noninteractive visualizations.

HC StratoMineR: A Web-Based Tool for the Rapid Analysis of High-Content Datasets.

High-content screening (HCS) can generate large multidimensional datasets and when aligned with the appropriate data mining tools, it can yield valuable insights into the mechanism of action of bioactive molecules. However, easy-to-use data mining tools are not widely available, with the result that these datasets are frequently underutilized. Here, we present HC StratoMineR, a web-based tool for high-content data analysis. It is a decision-supportive platform that guides even non-expert users through a high-content data analysis workflow. HC StratoMineR is built by using My Structured Query Language for storage and querying, PHP: Hypertext Preprocessor as the main programming language, and jQuery for additional user interface functionality. R is used for statistical calculations, logic and data visualizations. Furthermore, C++ and graphical processor unit power is diffusely embedded in R by using the rcpp and rpud libraries for operations that are computationally highly intensive. We show that we can use HC StratoMineR for the analysis of multivariate data from a high-content siRNA knock-down screen and a small-molecule screen. It can be used to rapidly filter out undesirable data; to select relevant data; and to perform quality control, data reduction, data exploration, morphological hit picking, and data clustering. Our results demonstrate that HC StratoMineR can be used to functionally categorize HCS hits and, thus, provide valuable information for hit prioritization.

An siRNA screen for ATG protein depletion reveals the extent of the unconventional functions of the autophagy proteome in virus replication.

Autophagy is a catabolic process regulated by the orchestrated action of the autophagy-related (ATG) proteins. Recent work indicates that some of the ATG proteins also have autophagy-independent roles. Using an unbiased siRNA screen approach, we explored the extent of these unconventional functions of ATG proteins. We determined the effects of the depletion of each ATG proteome component on the replication of six different viruses. Our screen reveals that up to 36% of the ATG proteins significantly alter the replication of at least one virus in an unconventional fashion. Detailed analysis of two candidates revealed an undocumented role for ATG13 and FIP200 in picornavirus replication that is independent of their function in autophagy as part of the ULK complex. The high numbers of unveiled ATG gene-specific and pathogen-specific functions of the ATG proteins calls for caution in the interpretation of data, which rely solely on the depletion of a single ATG protein to specifically ablate autophagy.

Screening of a Library of FDA-Approved Drugs Identifies Several Enterovirus Replication Inhibitors That Target Viral Protein 2C.

Enteroviruses (EVs) represent many important pathogens of humans. Unfortunately, no antiviral compounds currently exist to treat infections with these viruses. We screened the Prestwick Chemical Library, a library of approved drugs, for inhibitors of coxsackievirus B3, identified pirlindole as a potent novel inhibitor, and confirmed the inhibitory action of dibucaine, zuclopenthixol, fluoxetine, and formoterol. Upon testing of viruses of several EV species, we found that dibucaine and pirlindole inhibited EV-B and EV-D and that dibucaine also inhibited EV-A, but none of them inhibited EV-C or rhinoviruses (RVs). In contrast, formoterol inhibited all enteroviruses and rhinoviruses tested. All compounds acted through the inhibition of genome replication. Mutations in the coding sequence of the coxsackievirus B3 (CV-B3) 2C protein conferred resistance to dibucaine, pirlindole, and zuclopenthixol but not formoterol, suggesting that 2C is the target for this set of compounds. Importantly, dibucaine bound to CV-B3 protein 2C in vitro, whereas binding to a 2C protein carrying the resistance mutations was reduced, providing an explanation for how resistance is acquired.

The glucocorticoid mometasone furoate is a novel FXR ligand that decreases inflammatory but not metabolic gene expression.

The Farnesoid X receptor (FXR) regulates bile salt, glucose and cholesterol homeostasis by binding to DNA response elements, thereby activating gene expression (direct transactivation). FXR also inhibits the immune response via tethering to NF-κB (tethering transrepression). FXR activation therefore has therapeutic potential for liver and intestinal inflammatory diseases. We aim to identify and develop gene-selective FXR modulators, which repress inflammation, but do not interfere with its metabolic capacity. In a high-throughput reporter-based screen, mometasone furoate (MF) was identified as a compound that reduced NF-κB reporter activity in an FXR-dependent manner. MF reduced mRNA expression of pro-inflammatory cytokines, and induction of direct FXR target genes in HepG2-GFP-FXR cells and intestinal organoids was minor. Computational studies disclosed three putative binding modes of the compound within the ligand binding domain of the receptor. Interestingly, mutation of W469A residue within the FXR ligand binding domain abrogated the decrease in NF-κB activity. Finally, we show that MF-bound FXR inhibits NF-κB subunit p65 recruitment to the DNA of pro-inflammatory genes CXCL2 and IL8. Although MF is not suitable as selective anti-inflammatory FXR ligand due to nanomolar affinity for the glucocorticoid receptor, we show that separation between metabolic and anti-inflammatory functions of FXR can be achieved.