Search

High-throughput chemical screening approaches often employ microscopy to capture photomicrographs from multi-well cell culture plates, generating thousands of images that require time-consuming human analysis. To automate this subjective and time-consuming manual process, we have developed a method that uses deep learning to automatically classify digital assay images. We have trained a convolutional neural network (CNN) to perform binary and multi-class classification. The binary classifier binned assay images into healthy (comparable to untreated controls) and altered (not comparable to untreated-control) classes with >98% accuracy; the multi-class classifier assigned “Healthy,” “Intermediate” and “Altered” labels to assay images with >95% accuracy. Our dataset comprised high-resolution assay images from primary human hepatocytes and undifferentiated (proliferating) and differentiated 2D cultures of HepaRG cells. In this study we have focused on testing and fine-tuning various CNN architectures, including ResNet 34, 50 and 101. To visualize regions in the images that the CNN model used for classification, we employed Class Activation Maps (CAM). This allowed us to better understand the inner workings of the neural network and led to additional optimizations of the algorithm. The results indicate a strong correspondence between dosage and classifier-predicted scores, suggesting that these scores might be useful in further characterizing benchmark dose. Together, these results clearly demonstrate that deep-learning based automated image classification of cell morphology changes upon chemical-induced stress can yield highly accurate and reproducible assessments of cytotoxicity across a variety of cell types.

Background: In the screening phase of systematic review, researchers use detailed inclusion/exclusion criteria to decide whether each article in a set of candidate articles is relevant to the research question under consideration. A typical review may require screening thousands or tens of thousands of articles in and can utilize hundreds of person-hours of labor. Methods: Here we introduce SWIFT-Active Screener, a web-based, collaborative systematic review software application, designed to reduce the overall screening burden required during this resource-intensive phase of the review process. To prioritize articles for review, SWIFT-Active Screener uses active learning, a type of machine learning that incorporates user feedback during screening. Meanwhile, a negative binomial model is employed to estimate the number of relevant articles remaining in the unscreened document list. Using a simulation involving 26 diverse systematic review datasets that were previously screened by reviewers, we evaluated both the document prioritization and recall estimation methods. Results: On average, 95% of the relevant articles were identified after screening only 40% of the total reference list. In the 5 document sets with 5,000 or more references, 95% recall was achieved after screening only 34% of the available references, on average. Furthermore, the recall estimator we have proposed provides a useful, conservative estimate of the percentage of relevant documents identified during the screening process. Conclusion: SWIFT-Active Screener can result in significant time savings compared to traditional screening and the savings are increased for larger project sizes. Moreover, the integration of explicit recall estimation during screening solves an important challenge faced by all machine learning systems for document screening: when to stop screening a prioritized reference list. The software is currently available in the form of a multi-user, collaborative, online web application. Keywords: Systematic review, Evidence mapping, Active learning, Machine learning, Document screening, Recall estimation

Discovering temporal evolution of themes from a time-stamped collection of text poses a challenging statistical learning problem. Dynamic topic models offer a probabilistic modeling framework to decompose a corpus of text documents into “topics”, i.e., probability distributions over vocabulary terms, while simultaneously learning the temporal dynamics of the relative prevalence of these topics. We extend the dynamic topic model of Blei and Lafferty (2006) by fusing its multinomial factor model on topics with dynamic linear models that account for time trends and seasonality in topic prevalence. A Markov chain Monte Carlo (MCMC) algorithm that utilizes Polya-Gamma data augmentation is developed for posterior sampling. Conditional independencies in the model and sampling are made explicit, and our MCMC algorithm is parallelized where possible to allow for inference in large corpora. Our model and inference algorithm are validated with multiple synthetic examples, and we consider the applied problem of modeling trends in real estate listings from the housing website Zillow. We demonstrate in synthetic examples that sharing information across documents is critical for accurately estimating document-specific topic proportions. Analysis of the Zillow corpus demonstrates that the method is able to learn seasonal patterns and locally linear trends in topic prevalence.

BackgroundThe amount of published in vivo studies and the speed researchers are publishing them make it virtually impossible to follow the recent development in the field. Systematic review emerged as a method to summarise and analyse the studies quantitatively and critically but it is often out-of-date due to its lengthy process.MethodWe invited five machine learning and text-mining groups to build classifiers for identifying publications relevant to neuropathic pain (33814 training publications). We kept 1188 publications for the assessment of the performance of different classifiers. Two groups participated in the next stage: testing their algorithm on datasets labeled for psychosis (11777/2944) and datasets labeled for Vitamin D in multiple sclerosis (train/text: 2038/510).ResultThe performances (sensitive/specificity) of the most promising classifier built for neuropathic pain are: 95%/84%. The performance for psychosis and Vitamin D in multiple sclerosis datasets are 95%/73% and 100%/45%.ConclusionsMachine learning can significantly reduce the irrelevant publications in a systematic review, and save the scientists’ time and money. Classifier algorithms built for one dataset can be reapplied on another dataset in different field. We are building a machine learning service at the back of Systematic Review & Meta-analysis Facility (SyRF).

Changes in gene expression can help reveal the mechanisms of disease processes and the mode of action for toxicities and adverse effects on cellular responses induced by exposures to chemicals, drugs and environment agents. The U.S. Tox21 Federal collaboration, which currently quantifies the biological effects of nearly 10,000 chemicals via quantitative high-throughput screening(qHTS) in in vitro model systems, is now making an effort to incorporate gene expression profiling into the existing battery of assays. Whole transcriptome analyses performed on large numbers of samples using microarrays or RNA-Seq is currently cost-prohibitive. Accordingly, the Tox21 Program is pursuing a high-throughput transcriptomics (HTT) method that focuses on the targeted detection of gene expression for a carefully selected subset of the transcriptome that potentially can reduce the cost by a factor of 10-fold, allowing for the analysis of larger numbers of samples. To identify the optimal transcriptome subset, genes were sought that are (1) representative of the highly diverse biological space, (2) capable of serving as a proxy for expression changes in unmeasured genes, and (3) sufficient to provide coverage of well described biological pathways. A hybrid method for gene selection is presented herein that combines data-driven and knowledge-driven concepts into one cohesive method. Our approach is modular, applicable to any species, and facilitates a robust, quantitative evaluation of performance. In particular, we were able to perform gene selection such that the resulting set of “sentinel genes” adequately represents all known canonical pathways from Molecular Signature Database (MSigDB v4.0) and can be used to infer expression changes for the remainder of the transcriptome. The resulting computational model allowed us to choose a purely data-driven subset of 1500 sentinel genes, referred to as the S1500 set, which was then augmented using a knowledge-driven selection of additional genes to create the final S1500+ gene set. Our results indicate that the sentinel genes selected can be used to accurately predict pathway perturbations and biological relationships for samples under study.

Background Quality assessment of microarray data is an important and often challenging aspect of gene expression analysis. This task frequently involves the examination of a variety of summary statistics and diagnostic plots. The interpretation of these diagnostics is often subjective, and generally requires careful expert scrutiny. Results We show how an unsupervised classification technique based on the Expectation-Maximization (EM) algorithm and the naïve Bayes model can be used to automate microarray quality assessment. The method is flexible and can be easily adapted to accommodate alternate quality statistics and platforms. We evaluate our approach using Affymetrix 3' gene expression and exon arrays and compare the performance of this method to a similar supervised approach. Conclusion This research illustrates the efficacy of an unsupervised classification approach for the purpose of automated microarray data quality assessment. Since our approach requires only unannotated training data, it is easy to customize and to keep up-to-date as technology evolves. In contrast to other "black box" classification systems, this method also allows for intuitive explanations.

Recent research has demonstrated the utility of using supervised classification systems for automatic identification of low quality microarray data. However, this approach requires annotation of a large training set by a qualified expert. In this paper we demonstrate the utility of an unsupervised classification technique based on the Expectation-Maximization (EM) algorithm and naive Bayes classification. On our test set, this system exhibits performance comparable to that of an analogous supervised learner constructed from the same training data.

Eukaryotic marine microalgae like Dunaliella spp. have great potential as a feedstock for liquid transportation fuels because they grow fast and can accumulate high levels of triacylgycerides with little need for fresh water or land. Their growth rates vary between species and are dependent on environmental conditions. The cell cycle, starch and triacylglycerol accumulation are controlled by the diurnal light:dark cycle. Storage compounds like starch and triacylglycerol accumulate in the light when CO2 fixation rates exceed the need of assimilated carbon and energy for cell maintenance and division during the dark phase. To delineate environmental effects, we analyzed cell division rates, metabolism and transcriptional regulation in Dunaliella viridis in response to changes in light duration and growth temperatures. Its rate of cell division was increased under continuous light conditions, while a shift in temperature from 25 °C to 35 °C did not significantly affect the cell division rate, but increased the triacylglycerol content per cell several-fold under continuous light. The amount of saturated fatty acids in triacylglycerol fraction was more responsive to an increase in temperature than to a change in the light regime. Detailed fatty acid profiles showed that Dunaliella viridis incorporated lauric acid (C12:0) into triacylglycerol after 24 hours under continuous light. Transcriptome analysis identified potential regulators involved in the light and temperature-induced lipid accumulation in Dunaliella viridis.

The number and wide dynamic range of components found in biological matrixes present several challenges for global proteomics. In this perspective, we will examine the potential of zero-dimensional (0D), one-dimensional (1D), and two-dimensional (2D) separations coupled with Fourier-transform ion cyclotron resonance (FT-ICR) and time-of-flight (TOF) mass spectrometry (MS) for the analysis of complex mixtures. We describe and further develop previous reports on the space occupied by peptides, to calculate the theoretical peak capacity available to each separations-mass spectrometry method examined. Briefly, the peak capacity attainable by each of the mass analyzers was determined from the mass resolving power (RP) and the m/z space occupied by peptides considered from the mass distribution of tryptic peptides from National Center for Biotechnology Information's (NCBI's) nonredundant database. Our results indicate that reverse-phase-nanoHPLC (RP-nHPLC) separation coupled with FT-ICR MS offers an order of magnitude improvement in peak capacity over RP-nHPLC separation coupled with TOF MS. The addition of an orthogonal separation method, strong cation exchange (SCX), for 2D LC-MS demonstrates an additional 10-fold improvement in peak capacity over 1D LC-MS methods. Peak capacity calculations for 0D LC, two different 1D RP-HPLC methods, and 2D LC (with various numbers of SCX fractions) for both RP-HPLC methods coupled to FT-ICR and TOF MS are examined in detail. Peak capacity production rates, which take into account the total analysis time, are also considered for each of the methods. Furthermore, the significance of the space occupied by peptides is discussed.

Previous work investigating the performance of genetic algorithms (GAs) has attempted to develop a set of fitness landscapes, called “Royal Roads” functions, which should be ideally suited for search with GAs. Surprisingly, many studies have shown that genetic algorithms actually perform worse than random mutation hill-climbing on these landscapes, and several different explanations have been offered to account for these observations. Using a detailed stochastic model of genetic search on R1, we attempt to determine a lower bound for the required number of function evaluations, and then use it to evaluate the performance of an actual genetic algorithm on R1.

Laboratory dogs were vaccinated intramuscularly with a recombinant fusion protein (expressed and isolated from Escherichia coli) formulated with the Glaxo SmithKline Adjuvant System 02 (AS02). The fusion protein encoded Ac-MTP-1, a developmentally regulated astacinlike metalloprotease secreted by host-stimulated Ancylostoma caninum third-stage larvae (L3). Control dogs were injected intramuscularly with an equivalent amount of AS02 adjuvant alone. The vaccinated and control dogs were then challenged by s.c. injection of 500 L3 of the canine hookworm A. caninum. The vaccinated dogs developed prechallenge immunoglobulin G2 (IgG2) antibody responses specific to anti-Ac-MTP-1-fusion protein with titers ranging between 1:40,000 and 1:364,000, whereas they developed antigen-specific immunoglobulin E antibody responses with titers ranging between 1:500 and 1:1,500. By immunoblotting, canine sera obtained from the vaccinated dogs recognized a protein of the estimated apparent molecular weight of Ac-MTP-1 in activated L3 secretory products. Spearman rank order correlations between the canine intestinal adult hookworm burden and quantitative egg counts at necropsy and anti-Ac-MTP-1 IgG2 antibody titers revealed a statistically significant inverse association (r = -0.89; P = 0.04), suggesting that this molecule offers promise as a recombinant vaccine.

Background There is growing interest in using machine learning approaches to priority rank studies and reduce human burden in screening literature when conducting systematic reviews. In addition, identifying addressable questions during the problem formulation phase of systematic review can be challenging, especially for topics having a large literature base. Here, we assess the performance of the SWIFT-Review priority ranking algorithm for identifying studies relevant to a given research question. We also explore the use of SWIFT-Review during problem formulation to identify, categorize, and visualize research areas that are data rich/data poor within a large literature corpus. Methods Twenty case studies, including 15 public data sets, representing a range of complexity and size, were used to assess the priority ranking performance of SWIFT-Review. For each study, seed sets of manually annotated included and excluded titles and abstracts were used for machine training. The remaining references were then ranked for relevance using an algorithm that considers term frequency and latent Dirichlet allocation (LDA) topic modeling. This ranking was evaluated with respect to (1) the number of studies screened in order to identify 95 % of known relevant studies and (2) the “Work Saved over Sampling” (WSS) performance metric. To assess SWIFT-Review for use in problem formulation, PubMed literature search results for 171 chemicals implicated as EDCs were uploaded into SWIFT-Review (264,588 studies) and categorized based on evidence stream and health outcome. Patterns of search results were surveyed and visualized using a variety of interactive graphics. Results Compared with the reported performance of other tools using the same datasets, the SWIFT-Review ranking procedure obtained the highest scores on 11 out of 15 of the public datasets. Overall, these results suggest that using machine learning to triage documents for screening has the potential to save, on average, more than 50 % of the screening effort ordinarily required when using un-ordered document lists. In addition, the tagging and annotation capabilities of SWIFT-Review can be useful during the activities of scoping and problem formulation. Conclusions Text-mining and machine learning software such as SWIFT-Review can be valuable tools to reduce the human screening burden and assist in problem formulation. Electronic supplementary material The online version of this article (doi:10.1186/s13643-016-0263-z) contains supplementary material, which is available to authorized users.