A robust CETSA data analysis automation workflow for routine screening.

The Cellular Thermal Shift Assay (CETSA) enables the study of protein-ligand interactions in a cellular context. It provides valuable information on the binding affinity and specificity of both small and large molecule ligands in a relevant physiological context, hence forming a unique tool in drug discovery. Though high-throughput lab protocols exist for scaling up CETSA, subsequent data analysis and quality control remain laborious and limit experimental throughput. Here, we introduce a scalable and robust data analysis workflow which allows integration of CETSA into routine high throughput screening (HT-CETSA). This new workflow automates data analysis and incorporates quality control (QC), including outlier detection, sample and plate QC, and result triage. We describe the workflow and show its robustness against typical experimental artifacts, show scaling effects, and discuss the impact of data analysis automation by eliminating manual data processing steps.

High-throughput mechanistic screening of non-equilibrium inhibitors by a fully automated data analysis pipeline in early drug-discovery.

Recent efforts for increasing the success in drug discovery focus on an early, massive, and routine mechanistic and/or kinetic characterization of drug-target engagement as part of a design-make-test-analyze strategy. From an experimental perspective, many mechanistic assays can be translated into a scalable format on automation platforms and thereby enable routine characterization of hundreds or thousands of compounds. However, now the limiting factor to achieve such in-depth characterization at high-throughput becomes the quality-driven data analysis, the sheer scale of which outweighs the time available to the scientific staff of most labs. Therefore, automated analytical workflows are needed to enable such experimental scale-up. We have implemented such a fully automated workflow in Genedata Screener for time-dependent ligand-target binding analysis to characterize non-equilibrium inhibitors. The workflow automates Quality Control (QC) / data modelling and decision-making process in a staged analysis: (1) quality control of raw input data-fluorescence signal-based progress curves - featuring automated rejection of unsuitable measurements; (2) automated model selection - one-step versus two-step binding model - using statistical methods and biological validity rules; (3) result visualization in specific plots and annotated result tables, enabling the scientist to review large result sets efficiently and, at the same time, to rapidly identify and focus on interesting or unusual results; (4) an interactive user interface for immediate adjustment of automated decisions, where necessary. Applying this workflow to first-pass, high-throughput kinetic studies on kinase projects has allowed us to surmount previously rate-limiting manual analysis steps and boost productivity; and is now routinely embedded in a biopharma discovery research process.

An efficient and scalable data analysis solution for automated electrophysiology platforms.

Ion channels are drug targets for neurologic, cardiac, and immunologic diseases. Many disease-associated mutations and drugs modulate voltage-gated ion channel activation and inactivation, suggesting that characterizing state-dependent effects of test compounds at an early stage of drug development can be of great benefit. Historically, the effects of compounds on ion channel biophysical properties and voltage-dependent activation/inactivation could only be assessed by using low-throughput, manual patch clamp recording techniques. In recent years, automated patch clamp (APC) platforms have drastically increased in throughput. In contrast to their broad utilization in compound screening, APC platforms have rarely been used for mechanism of action studies, in large part due to the lack of sophisticated, scalable analysis methods for processing the large amount of data generated by APC platforms. In the current study, we developed a highly efficient and scalable software workflow to overcome this challenge. This method, to our knowledge the first of its kind, enables automated curve fitting and complex analysis of compound effects. Using voltage-gated sodium channels as an example, we were able to immediately assess the effects of test compounds on a spectrum of biophysical properties, including peak current, voltage-dependent steady state activation/inactivation, and time constants of activation and fast inactivation. Overall, this automated data analysis method provides a novel solution for in-depth analysis of large-scale APC data, and thus will significantly impact ion channel research and drug discovery.

Benchmarking feature selection methods for compressing image information in high-content screening.

Biopharmaceutical drug discovery, as of today is a highly automated, high throughput endeavor, where many screening technologies produce a high-dimensional measurement per sample. A striking example is High Content Screening (HCS), which utilizes automated microscopy to systematically access the wealth of information contained in biological assays. Exploiting HCS to its full potential traditionally requires extracting a high number of features from the images to capture as much information as possible, then performing algorithmic analysis and complex data visualization in order to render this high-dimensional data into an interpretable and instructive information for guiding drug development. In this process, automated feature selection methods condense the feature set to reduce non-useful or redundant information and render it more meaningful. We compare 12 state-of-the-art feature selection methods (both supervised and unsupervised) by systematically testing them on two HCS datasets from drug screening imaging assays of high practical relevance. Considering as evaluation metrics standard plate-, assay- or compound statistics on the final results, we assess the generalizability and importance of the selected features by use of automated machine learning (AutoML) to achieve an unbiased evaluation across methods. Results provide practical guidance on experiment design, optimal sizing of a reduced feature set and choice of feature selection method, both in situations where useful experimental control states are available (enabling use of supervised algorithms) or where such controls are unavailable, using unsupervised techniques.

Deep Learning-Based HCS Image Analysis for the Enterprise.

Drug discovery programs are moving increasingly toward phenotypic imaging assays to model disease-relevant pathways and phenotypes in vitro. These assays offer richer information than target-optimized assays by investigating multiple cellular pathways simultaneously and producing multiplexed readouts. However, extracting the desired information from complex image data poses significant challenges, preventing broad adoption of more sophisticated phenotypic assays. Deep learning-based image analysis can address these challenges by reducing the effort required to analyze large volumes of complex image data at a quality and speed adequate for routine phenotypic screening in pharmaceutical research. However, while general purpose deep learning frameworks are readily available, they are not readily applicable to images from automated microscopy. During the past 3 years, we have optimized deep learning networks for this type of data and validated the approach across diverse assays with several industry partners. From this work, we have extracted five essential design principles that we believe should guide deep learning-based analysis of high-content images and multiparameter data: (1) insightful data representation, (2) automation of training, (3) multilevel quality control, (4) knowledge embedding and transfer to new assays, and (5) enterprise integration. We report a new deep learning-based software that embodies these principles, Genedata Imagence, which allows screening scientists to reliably detect stable endpoints for primary drug response, assess toxicity and safety-relevant effects, and discover new phenotypes and compound classes. Furthermore, we show how the software retains expert knowledge from its training on a particular assay and successfully reapplies it to different, novel assays in an automated fashion.

Considerations for improved performance of competition association assays analysed with the Motulsky-Mahan's "kinetics of competitive binding" model.

BACKGROUND AND PURPOSE: Target engagement dynamics can influence drugs' pharmacological effects. Kinetic parameters for drug:target interactions are often quantified by evaluating competition association experiments-measuring simultaneous protein binding of labelled tracers and unlabelled test compounds over time-with Motulsky-Mahan's "kinetics of competitive binding" model. Despite recent technical improvements, the current assay formats impose practical limitations to this approach. This study aims at the characterisation, understanding and prevention of these experimental constraints, and associated analytical challenges. EXPERIMENTAL APPROACH: Monte Carlo simulations were used to run virtual kinetic and equilibrium tracer binding and competition experiments in both normal and perturbed assay conditions. Data were fitted to standard equations derived from the mass action law (including Motulsky-Mahan's) and to extended versions aiming to cope with frequently observed deviations of the canonical traces. Results were compared to assess the precision and accuracy of these models and identify experimental factors influencing their performance. KEY RESULTS: Key factors influencing the precision and accuracy of the Motulsky-Mahan model are the interplay between compound dissociation rates, measurement time and interval frequency, tracer concentration and binding kinetics and the relative abundance of equilibrium complexes in vehicle controls. Experimental results produced recommendations for better design of tracer characterisation experiments and new strategies to deal with systematic signal decay. CONCLUSIONS AND IMPLICATIONS: Our data advances our comprehension of the Motulsky-Mahan kinetics of competitive binding models and provides experimental design recommendations, data analysis tools, and general guidelines for its practical application to in vitro pharmacology and drug screening.

Know When You Don't Know: A Robust Deep Learning Approach in the Presence of Unknown Phenotypes.

Deep convolutional neural networks show outstanding performance in image-based phenotype classification given that all existing phenotypes are presented during the training of the network. However, in real-world high-content screening (HCS) experiments, it is often impossible to know all phenotypes in advance. Moreover, novel phenotype discovery itself can be an HCS outcome of interest. This aspect of HCS is not yet covered by classical deep learning approaches. When presenting an image with a novel phenotype to a trained network, it fails to indicate a novelty discovery but assigns the image to a wrong phenotype. To tackle this problem and address the need for novelty detection, we use a recently developed Bayesian approach for deep neural networks called Monte Carlo (MC) dropout to define different uncertainty measures for each phenotype prediction. With real HCS data, we show that these uncertainty measures allow us to identify novel or unclear phenotypes. In addition, we also found that the MC dropout method results in a significant improvement of classification accuracy. The proposed procedure used in our HCS case study can be easily transferred to any existing network architecture and will be beneficial in terms of accuracy and novelty detection.

Developing Deep Learning Applications for Life Science and Pharma Industry.

Deep Learning has boosted artificial intelligence over the past 5 years and is seen now as one of the major technological innovation areas, predicted to replace lots of repetitive, but complex tasks of human labor within the next decade. It is also expected to be 'game changing' for research activities in pharma and life sciences, where large sets of similar yet complex data samples are systematically analyzed. Deep learning is currently conquering formerly expert domains especially in areas requiring perception, previously not amenable to standard machine learning. A typical example is the automated analysis of images which are typically produced en-masse in many domains, e. g., in high-content screening or digital pathology. Deep learning enables to create competitive applications in so-far defined core domains of 'human intelligence'. Applications of artificial intelligence have been enabled in recent years by (i) the massive availability of data samples, collected in pharma driven drug programs (='big data') as well as (ii) deep learning algorithmic advancements and (iii) increase in compute power. Such applications are based on software frameworks with specific strengths and weaknesses. Here, we introduce typical applications and underlying frameworks for deep learning with a set of practical criteria for developing production ready solutions in life science and pharma research. Based on our own experience in successfully developing deep learning applications we provide suggestions and a baseline for selecting the most suited frameworks for a future-proof and cost-effective development.

Unified Software Solution for Efficient SPR Data Analysis in Drug Research.

Surface plasmon resonance (SPR) is a powerful method for obtaining detailed molecular interaction parameters. Modern instrumentation with its increased throughput has enabled routine screening by SPR in hit-to-lead and lead optimization programs, and SPR has become a mainstream drug discovery technology. However, the processing and reporting of SPR data in drug discovery are typically performed manually, which is both time-consuming and tedious. Here, we present the workflow concept, design and experiences with a software module relying on a single, browser-based software platform for the processing, analysis, and reporting of SPR data. The efficiency of this concept lies in the immediate availability of end results: data are processed and analyzed upon loading the raw data file, allowing the user to immediately quality control the results. Once completed, the user can automatically report those results to data repositories for corporate access and quickly generate printed reports or documents. The software module has resulted in a very efficient and effective workflow through saved time and improved quality control. We discuss these benefits and show how this process defines a new benchmark in the drug discovery industry for the handling, interpretation, visualization, and sharing of SPR data.

Genomic organization of eukaryotic tRNAs.

BACKGROUND: Surprisingly little is known about the organization and distribution of tRNA genes and tRNA-related sequences on a genome-wide scale. While tRNA gene complements are usually reported in passing as part of genome annotation efforts, and peculiar features such as the tandem arrangements of tRNA gene in Entamoeba histolytica have been described in some detail, systematic comparative studies are rare and mostly restricted to bacteria. We therefore set out to survey the genomic arrangement of tRNA genes and pseudogenes in a wide range of eukaryotes to identify common patterns and taxon-specific peculiarities. RESULTS: In line with previous reports, we find that tRNA complements evolve rapidly and tRNA gene and pseudogene locations are subject to rapid turnover. At phylum level, the distributions of the number of tRNA genes and pseudogenes numbers are very broad, with standard deviations on the order of the mean. Even among closely related species we observe dramatic changes in local organization. For instance, 65% and 87% of the tRNA genes and pseudogenes are located in genomic clusters in zebrafish and stickleback, resp., while such arrangements are relatively rare in the other three sequenced teleost fish genomes. Among basal metazoa, Trichoplax adherens has hardly any duplicated tRNA gene, while the sea anemone Nematostella vectensis boasts more than 17000 tRNA genes and pseudogenes. Dramatic variations are observed even within the eutherian mammals. Higher primates, for instance, have 616 +/- 120 tRNA genes and pseudogenes of which 17% to 36% are arranged in clusters, while the genome of the bushbaby Otolemur garnetti has 45225 tRNA genes and pseudogenes of which only 5.6% appear in clusters. In contrast, the distribution is surprisingly uniform across plant genomes. Consistent with this variability, syntenic conservation of tRNA genes and pseudogenes is also poor in general, with turn-over rates comparable to those of unconstrained sequence elements. Despite this large variation in abundance in Eukarya we observe a significant correlation between the number of tRNA genes, tRNA pseudogenes, and genome size. CONCLUSIONS: The genomic organization of tRNA genes and pseudogenes shows complex lineage-specific patterns characterized by an extensive variability that is in striking contrast to the extreme levels of sequence-conservation of the tRNAs themselves. The comprehensive analysis of the genomic organization of tRNA genes and pseudogenes in Eukarya provides a basis for further studies into the interplay of tRNA gene arrangements and genome organization in general.

Noncoding RNA of glutamine synthetase I modulates antibiotic production in Streptomyces coelicolor A3(2).

Overexpression of antisense chromosomal cis-encoded noncoding RNAss (ncRNAs) in glutamine synthetase I resulted in a decrease in growth, protein synthesis, and antibiotic production in Streptomyces coelicolor. In addition, we predicted 3,597 cis-encoded ncRNAs and validated 13 of them experimentally, including several ncRNAs that are differentially expressed in bacterial hormone-defective mutants.

Comparative analysis of structured RNAs in S. cerevisiae indicates a multitude of different functions.

BACKGROUND: Non-coding RNAs (ncRNAs) are an emerging focus for both computational analysis and experimental research, resulting in a growing number of novel, non-protein coding transcripts with often unknown functions. Whole genome screens in higher eukaryotes, for example, provided evidence for a surprisingly large number of ncRNAs. To supplement these searches, we performed a computational analysis of seven yeast species and searched for new ncRNAs and RNA motifs. RESULTS: A comparative analysis of the genomes of seven yeast species yielded roughly 2800 genomic loci that showed the hallmarks of evolutionary conserved RNA secondary structures. A total of 74% of these regions overlapped with annotated non-coding or coding genes in yeast. Coding sequences that carry predicted structured RNA elements belong to a limited number of groups with common functions, suggesting that these RNA elements are involved in post-transcriptional regulation and/or cellular localization. About 700 conserved RNA structures were found outside annotated coding sequences and known ncRNA genes. Many of these predicted elements overlapped with UTR regions of particular classes of protein coding genes. In addition, a number of RNA elements overlapped with previously characterized antisense transcripts. Transcription of about 120 predicted elements located in promoter regions and other, previously un-annotated, intergenic regions was supported by tiling array experiments, ESTs, or SAGE data. CONCLUSION: Our computational predictions strongly suggest that yeasts harbor a substantial pool of several hundred novel ncRNAs. In addition, we describe a large number of RNA structures in coding sequences and also within antisense transcripts that were previously characterized using tiling arrays.

On the maximal cliques in c-max-tolerance graphs and their application in clustering molecular sequences.

Given a set S of n locally aligned sequences, it is a needed prerequisite to partition it into groups of very similar sequences to facilitate subsequent computations, such as the generation of a phylogenetic tree. This article introduces a new method of clustering which partitions S into subsets such that the overlap of each pair of sequences within a subset is at least a given percentage c of the lengths of the two sequences. We show that this problem can be reduced to finding all maximal cliques in a special kind of max-tolerance graph which we call a c-max-tolerance graph. Previously we have shown that finding all maximal cliques in general max-tolerance graphs can be done efficiently in O(n3 + out). Here, using a new kind of sweep-line algorithm, we show that the restriction to c-max-tolerance graphs yields a better runtime of O(n2 log n + out). Furthermore, we present another algorithm which is much easier to implement, and though theoretically slower than the first one, is still running in polynomial time. We then experimentally analyze the number and structure of all maximal cliques in a c-max-tolerance graph, depending on the chosen c-value. We apply our simple algorithm to artificial and biological data and we show that this implementation is much faster than the well-known application Cliquer. By introducing a new heuristic that uses the set of all maximal cliques to partition S, we finally show that the computed partition gives a reasonable clustering for biological data sets.

Open reading frames provide a rich pool of potential natural antisense transcripts in fungal genomes.

Natural antisense transcripts are reported from all kingdoms of life and several recent reports of genomewide screens indicate that they are widely distributed. These transcripts seem to be involved in various biological functions and may govern the expression of their respective sense partner. Very little, however, is known about the degree of evolutionary conservation of antisense transcripts. Furthermore, none of the earlier analyses has studied whether antisense relationships are solely dual or involved in more complex relationships. Here we present a systematic screen for cis- and trans-located antisense transcripts based on open reading frames (ORFs) from five fungal species. The relative number of ORFs involved in antisense relationships varies greatly between the five species. In addition, other significant differences are found between the species, such as the mean length of the antisense region. The majority of trans-located antisense transcripts is found to be involved in complex relationships, resulting in highly connected networks. The analysis of the degree of evolutionary conservation of antisense transcripts shows that most antisense transcripts have no ortholog in any other species. An annotation of antisense transcripts based on Gene Ontology directs to common terms and shows that proteins of genes involved in antisense relationships preferentially localize to the nucleus with common functions in the regulation or maintenance of nucleic acids.

Regional patterns of gene expression in human and chimpanzee brains.

We have analyzed gene expression in various brain regions of humans and chimpanzees. Within both human and chimpanzee individuals, the transcriptomes of the cerebral cortex are very similar to each other and differ more between individuals than among regions within an individual. In contrast, the transcriptomes of the cerebral cortex, the caudate nucleus, and the cerebellum differ substantially from each other. Between humans and chimpanzees, 10% of genes differ in their expression in at least one region of the brain. The majority of these expression differences are shared among all brain regions. Whereas genes encoding proteins involved in signal transduction and cell differentiation differ significantly between brain regions within individuals, no such pattern is seen between the species. However, a subset of genes that show expression differences between humans and chimpanzees are distributed nonrandomly across the genome. Furthermore, genes that show an elevated expression level in humans are statistically significantly enriched in regions that are recently duplicated in humans.