Drug target prediction through deep learning functional representation of gene signatures.

Many machine learning applications in bioinformatics currently rely on matching gene identities when analyzing input gene signatures and fail to take advantage of preexisting knowledge about gene functions. To further enable comparative analysis of OMICS datasets, including target deconvolution and mechanism of action studies, we develop an approach that represents gene signatures projected onto their biological functions, instead of their identities, similar to how the word2vec technique works in natural language processing. We develop the Functional Representation of Gene Signatures (FRoGS) approach by training a deep learning model and demonstrate that its application to the Broad Institute's L1000 datasets results in more effective compound-target predictions than models based on gene identities alone. By integrating additional pharmacological activity data sources, FRoGS significantly increases the number of high-quality compound-target predictions relative to existing approaches, many of which are supported by in silico and/or experimental evidence. These results underscore the general utility of FRoGS in machine learning-based bioinformatics applications. Prediction networks pre-equipped with the knowledge of gene functions may help uncover new relationships among gene signatures acquired by large-scale OMICs studies on compounds, cell types, disease models, and patient cohorts.

Predicting drug approvals: The Novartis data science and artificial intelligence challenge.

We describe a novel collaboration between academia and industry, an in-house data science and artificial intelligence challenge held by Novartis to develop machine-learning models for predicting drug-development outcomes, building upon research at MIT using data from Informa as the starting point. With over 50 cross-functional teams from 25 Novartis offices around the world participating in the challenge, the domain expertise of these Novartis researchers was leveraged to create predictive models with greater sophistication. Ultimately, two winning teams developed models that outperformed the baseline MIT model-areas under the curve of 0.88 and 0.84 versus 0.78, respectively-through state-of-the-art machine-learning algorithms and the use of newly incorporated features and data. In addition to validating the variables shown to be associated with drug approval in the earlier MIT study, the challenge also provided new insights into the drivers of drug-development success and failure.

Targeting p130Cas- and microtubule-dependent MYC regulation sensitizes pancreatic cancer to ERK MAPK inhibition.

To identify therapeutic targets for KRAS mutant pancreatic cancer, we conduct a druggable genome small interfering RNA (siRNA) screen and determine that suppression of BCAR1 sensitizes pancreatic cancer cells to ERK inhibition. Integrative analysis of genome-scale CRISPR-Cas9 screens also identify BCAR1 as a top synthetic lethal interactor with mutant KRAS. BCAR1 encodes the SRC substrate p130Cas. We determine that SRC-inhibitor-mediated suppression of p130Cas phosphorylation impairs MYC transcription through a DOCK1-RAC1-ß-catenin-dependent mechanism. Additionally, genetic suppression of TUBB3, encoding the ßIII-tubulin subunit of microtubules, or pharmacological inhibition of microtubule function decreases levels of MYC protein in a calpain-dependent manner and potently sensitizes pancreatic cancer cells to ERK inhibition. Accordingly, the combination of a dual SRC/tubulin inhibitor with an ERK inhibitor cooperates to reduce MYC protein and synergistically suppress the growth of KRAS mutant pancreatic cancer. Thus, we demonstrate that mechanistically diverse combinations with ERK inhibition suppress MYC to impair pancreatic cancer proliferation.

Proteomic Analysis of Non-human Primate Peripheral Blood Mononuclear Cells During Burkholderia mallei Infection Reveals a Role of Ezrin in Glanders Pathogenesis.

Burkholderia mallei, the causative agent of glanders, is a gram-negative intracellular bacterium. Depending on different routes of infection, the disease is manifested by pneumonia, septicemia, and chronic infections of the skin. B. mallei poses a serious biological threat due to its ability to infect via aerosol route, resistance to multiple antibiotics and to date there are no US Food and Drug Administration (FDA) approved vaccines available. Induction of innate immunity, inflammatory cytokines and chemokines following B. mallei infection, have been observed in in vitro and small rodent models; however, a global characterization of host responses has never been systematically investigated using a non-human primate (NHP) model. Here, using a liquid chromatography-tandem mass spectrometry (LC-MS/MS) approach, we identified alterations in expression levels of host proteins in peripheral blood mononuclear cells (PBMCs) originating from naïve rhesus macaques (Macaca mulatta), African green monkeys (Chlorocebus sabaeus), and cynomolgus macaques (Macaca fascicularis) exposed to aerosolized B. mallei. Gene ontology (GO) analysis identified several statistically significant overrepresented biological annotations including complement and coagulation cascade, nucleoside metabolic process, vesicle-mediated transport, intracellular signal transduction and cytoskeletal protein binding. By integrating an LC-MS/MS derived proteomics dataset with a previously published B. mallei host-pathogen interaction dataset, a statistically significant predictive protein-protein interaction (PPI) network was constructed. Pharmacological perturbation of one component of the PPI network, specifically ezrin, reduced B. mallei mediated interleukin-1ß (IL-1ß). On the contrary, the expression of IL-1ß receptor antagonist (IL-1Ra) was upregulated upon pretreatment with the ezrin inhibitor. Taken together, inflammasome activation as demonstrated by IL-1ß production and the homeostasis of inflammatory response is critical during the pathogenesis of glanders. Furthermore, the topology of the network reflects the underlying molecular mechanism of B. mallei infections in the NHP model.

A cell-level quality control workflow for high-throughput image analysis.

BACKGROUND: Image-based high throughput (HT) screening provides a rich source of information on dynamic cellular response to external perturbations. The large quantity of data generated necessitates computer-aided quality control (QC) methodologies to flag imaging and staining artifacts. Existing image- or patch-level QC methods require separate thresholds to be simultaneously tuned for each image quality metric used, and also struggle to distinguish between artifacts and valid cellular phenotypes. As a result, extensive time and effort must be spent on per-assay QC feature thresholding, and valid images and phenotypes may be discarded while image- and cell-level artifacts go undetected. RESULTS: We present a novel cell-level QC workflow built on machine learning approaches for classifying artifacts from HT image data. First, a phenotype sampler based on unlabeled clustering collects a comprehensive subset of cellular phenotypes, requiring only the inspection of a handful of images per phenotype for validity. A set of one-class support vector machines are then trained on each biologically valid image phenotype, and used to classify individual objects in each image as valid cells or artifacts. We apply this workflow to two real-world large-scale HT image datasets and observe that the ratio of artifact to total object area (ARcell) provides a single robust assessment of image quality regardless of the underlying causes of quality issues. Gating on this single intuitive metric, partially contaminated images can be salvaged and highly contaminated images can be excluded before image-level phenotype summary, enabling a more reliable characterization of cellular response dynamics. CONCLUSIONS: Our cell-level QC workflow enables identification of artificial cells created not only by staining or imaging artifacts but also by the limitations of image segmentation algorithms. The single readout ARcell that summaries the ratio of artifacts contained in each image can be used to reliably rank images by quality and more accurately determine QC cutoff thresholds. Machine learning-based cellular phenotype clustering and sampling reduces the amount of manual work required for training example collection. Our QC workflow automatically handles assay-specific phenotypic variations and generalizes to different HT image assays.

Dexamethasone inhibits respiratory syncytial virus-driven mucus production while increasing viral replication without altering antiviral interferon signaling.

Respiratory syncytial virus (RSV) infection can cause mucus overproduction and bronchiolitis in infants leading to severe disease and hospitalization. As a therapeutic strategy, immune modulatory agents may help prevent RSV-driven immune responses that cause severe airway disease. We developed a high throughput screen to identify compounds that reduced RSV-driven mucin 5AC (Muc5AC) expression and identified dexamethasone. Despite leading to a pronounced reduction in RSV-driven Muc5AC, dexamethasone increased RSV infection in vitro and delayed viral clearance in mice. This correlated with reduced expression of a subset of immune response genes and reduced lymphocyte infiltration in vivo. Interestingly, dexamethasone increased RSV infection levels without altering antiviral interferon signaling. In summary, the immunosuppressive activities of dexamethasone had favorable inhibitory effects on RSV-driven mucus production yet prevented immune defense activities that limit RSV infection in vitro and in vivo. These findings offer an explanation for the lack of efficacy of glucocorticoids in RSV-infected patients.

Metascape provides a biologist-oriented resource for the analysis of systems-level datasets.

A critical component in the interpretation of systems-level studies is the inference of enriched biological pathways and protein complexes contained within OMICs datasets. Successful analysis requires the integration of a broad set of current biological databases and the application of a robust analytical pipeline to produce readily interpretable results. Metascape is a web-based portal designed to provide a comprehensive gene list annotation and analysis resource for experimental biologists. In terms of design features, Metascape combines functional enrichment, interactome analysis, gene annotation, and membership search to leverage over 40 independent knowledgebases within one integrated portal. Additionally, it facilitates comparative analyses of datasets across multiple independent and orthogonal experiments. Metascape provides a significantly simplified user experience through a one-click Express Analysis interface to generate interpretable outputs. Taken together, Metascape is an effective and efficient tool for experimental biologists to comprehensively analyze and interpret OMICs-based studies in the big data era.

Identification of a Xist silencing domain by Tiling CRISPR.

Despite essential roles played by long noncoding RNAs (lncRNAs) in development and disease, methods to determine lncRNA cis-elements are lacking. Here, we developed a screening method named "Tiling CRISPR" to identify lncRNA functional domains. Using this approach, we identified Xist A-Repeats as the silencing domain, an observation in agreement with published work, suggesting Tiling CRISPR feasibility. Mechanistic analysis suggested a novel function for Xist A-repeats in promoting Xist transcription. Overall, our method allows mapping of lncRNA functional domains in an unbiased and potentially high-throughput manner to facilitate the understanding of lncRNA functions.

Open-source discovery of chemical leads for next-generation chemoprotective antimalarials.

To discover leads for next-generation chemoprotective antimalarial drugs, we tested more than 500,000 compounds for their ability to inhibit liver-stage development of luciferase-expressing Plasmodium spp. parasites (681 compounds showed a half-maximal inhibitory concentration of less than 1 micromolar). Cluster analysis identified potent and previously unreported scaffold families as well as other series previously associated with chemoprophylaxis. Further testing through multiple phenotypic assays that predict stage-specific and multispecies antimalarial activity distinguished compound classes that are likely to provide symptomatic relief by reducing asexual blood-stage parasitemia from those which are likely to only prevent malaria. Target identification by using functional assays, in vitro evolution, or metabolic profiling revealed 58 mitochondrial inhibitors but also many chemotypes possibly with previously unidentified mechanisms of action.

A Fully Automated High-Throughput Flow Cytometry Screening System Enabling Phenotypic Drug Discovery.

The goal of high-throughput screening is to enable screening of compound libraries in an automated manner to identify quality starting points for optimization. This often involves screening a large diversity of compounds in an assay that preserves a connection to the disease pathology. Phenotypic screening is a powerful tool for drug identification, in that assays can be run without prior understanding of the target and with primary cells that closely mimic the therapeutic setting. Advanced automation and high-content imaging have enabled many complex assays, but these are still relatively slow and low throughput. To address this limitation, we have developed an automated workflow that is dedicated to processing complex phenotypic assays for flow cytometry. The system can achieve a throughput of 50,000 wells per day, resulting in a fully automated platform that enables robust phenotypic drug discovery. Over the past 5 years, this screening system has been used for a variety of drug discovery programs, across many disease areas, with many molecules advancing quickly into preclinical development and into the clinic. This report will highlight a diversity of approaches that automated flow cytometry has enabled for phenotypic drug discovery.

Mapping the malaria parasite druggable genome by using in vitro evolution and chemogenomics.

Chemogenetic characterization through in vitro evolution combined with whole-genome analysis can identify antimalarial drug targets and drug-resistance genes. We performed a genome analysis of 262 Plasmodium falciparum parasites resistant to 37 diverse compounds. We found 159 gene amplifications and 148 nonsynonymous changes in 83 genes associated with drug-resistance acquisition, where gene amplifications contributed to one-third of resistance acquisition events. Beyond confirming previously identified multidrug-resistance mechanisms, we discovered hitherto unrecognized drug target-inhibitor pairs, including thymidylate synthase and a benzoquinazolinone, farnesyltransferase and a pyrimidinedione, and a dipeptidylpeptidase and an arylurea. This exploration of the P. falciparum resistome and druggable genome will likely guide drug discovery and structural biology efforts, while also advancing our understanding of resistance mechanisms available to the malaria parasite.

Systems-based analysis of RIG-I-dependent signalling identifies KHSRP as an inhibitor of RIG-I receptor activation.

Retinoic acid-inducible gene I (RIG-I) receptor recognizes 5'-triphosphorylated RNA and triggers a signalling cascade that results in the induction of type-I interferon (IFN)-dependent responses. Its precise regulation represents a pivotal balance between antiviral defences and autoimmunity. To elucidate the cellular cofactors that regulate RIG-I signalling, we performed two global RNA interference analyses to identify both positive and negative regulatory nodes operating on the signalling pathway during virus infection. These factors were integrated with experimentally and computationally derived interactome data to build a RIG-I protein interaction network. Our analysis revealed diverse cellular processes, including the unfolded protein response, Wnt signalling and RNA metabolism, as critical cellular components governing innate responses to non-self RNA species. Importantly, we identified K-Homology Splicing Regulatory Protein (KHSRP) as a negative regulator of this pathway. We find that KHSRP associates with the regulatory domain of RIG-I to maintain the receptor in an inactive state and attenuate its sensing of viral RNA (vRNA). Consistent with increased RIG-I antiviral signalling in the absence of KHSRP, viral replication is reduced when KHSRP expression is knocked down both in vitro and in vivo. Taken together, these data indicate that KHSRP functions as a checkpoint regulator of the innate immune response to pathogen challenge.

An Automatic Quality Control Pipeline for High-Throughput Screening Hit Identification.

The correction or removal of signal errors in high-throughput screening (HTS) data is critical to the identification of high-quality lead candidates. Although a number of strategies have been previously developed to correct systematic errors and to remove screening artifacts, they are not universally effective and still require fair amount of human intervention. We introduce a fully automated quality control (QC) pipeline that can correct generic interplate systematic errors and remove intraplate random artifacts. The new pipeline was first applied to ~100 large-scale historical HTS assays; in silico analysis showed auto-QC led to a noticeably stronger structure-activity relationship. The method was further tested in several independent HTS runs, where QC results were sampled for experimental validation. Significantly increased hit confirmation rates were obtained after the QC steps, confirming that the proposed method was effective in enriching true-positive hits. An implementation of the algorithm is available to the screening community.

A broad analysis of resistance development in the malaria parasite.

Microbial resistance to chemotherapy has caused countless deaths where malaria is endemic. Chemotherapy may fail either due to pre-existing resistance or evolution of drug-resistant parasites. Here we use a diverse set of antimalarial compounds to investigate the acquisition of drug resistance and the degree of cross-resistance against common resistance alleles. We assess cross-resistance using a set of 15 parasite lines carrying resistance-conferring alleles in pfatp4, cytochrome bc1, pfcarl, pfdhod, pfcrt, pfmdr, pfdhfr, cytoplasmic prolyl t-RNA synthetase or hsp90. Subsequently, we assess whether resistant parasites can be obtained after several rounds of drug selection. Twenty-three of the 48 in vitro selections result in resistant parasites, with time to resistance onset ranging from 15 to 300 days. Our data indicate that pre-existing resistance may not be a major hurdle for novel-target antimalarial candidates, and focusing our attention on fast-killing compounds may result in a slower onset of clinical resistance.

High-Throughput Assay and Discovery of Small Molecules that Interrupt Malaria Transmission.

Preventing transmission is an important element of malaria control. However, most of the current available methods to assay for malaria transmission blocking are relatively low throughput and cannot be applied to large chemical libraries. We have developed a high-throughput and cost-effective assay, the Saponin-lysis Sexual Stage Assay (SaLSSA), for identifying small molecules with transmission-blocking capacity. SaLSSA analysis of 13,983 unique compounds uncovered that >90% of well-characterized antimalarials, including endoperoxides and 4-aminoquinolines, as well as compounds active against asexual blood stages, lost most of their killing activity when parasites developed into metabolically quiescent stage V gametocytes. On the other hand, we identified compounds with consistent low nanomolar transmission-blocking activity, some of which showed cross-reactivity against asexual blood and liver stages. The data clearly emphasize substantial physiological differences between sexual and asexual parasites and provide a tool and starting points for the discovery and development of transmission-blocking drugs.

Meta- and Orthogonal Integration of Influenza "OMICs" Data Defines a Role for UBR4 in Virus Budding.

Several systems-level datasets designed to dissect host-pathogen interactions during influenza A infection have been reported. However, apparent discordance among these data has hampered their full utility toward advancing mechanistic and therapeutic knowledge. To collectively reconcile these datasets, we performed a meta-analysis of data from eight published RNAi screens and integrated these data with three protein interaction datasets, including one generated within the context of this study. Further integration of these data with global virus-host interaction analyses revealed a functionally validated biochemical landscape of the influenza-host interface, which can be queried through a simplified and customizable web portal (http://www.metascape.org/IAV). Follow-up studies revealed that the putative ubiquitin ligase UBR4 associates with the viral M2 protein and promotes apical transport of viral proteins. Taken together, the integrative analysis of influenza OMICs datasets illuminates a viral-host network of high-confidence human proteins that are essential for influenza A virus replication.

Alveolar macrophages infected with Ames or Sterne strain of Bacillus anthracis elicit differential molecular expression patterns.

Alveolar macrophages (AMs) phagocytose Bacillus anthracis following inhalation and induce the production of pro-inflammatory cytokines and chemokines to mediate the activation of innate immunity. Ames, the virulent strain of B. anthracis, contains two plasmids that encode the antiphagocytic poly-γ-d-glutamic acid capsule and the lethal toxin. The attenuated Sterne strain of B. anthracis, which lacks the plasmid encoding capsule, is widely adapted as a vaccine strain. Although differences in the outcome of infection with the two strains may have originated from the presence or absence of an anti-phagocytic capsule, the disease pathogenesis following infection will be manifested via the host responses, which is not well understood. To gain understanding of the host responses at cellular level, a microarray analysis was performed using primary rhesus macaque AMs infected with either Ames or Sterne spores. Notably, 528 human orthologs were identified to be differentially expressed in AMs infected with either strain of the B. anthracis. Meta-analyses revealed genes differentially expressed in response to B. anthracis infection were also induced upon infections with multiple pathogens such as Francisella Novicida or Staphylococcus aureus. This suggests the existence of a common molecular signature in response to pathogen infections. Importantly, the microarray and protein expression data for certain cytokines, chemokines and host factors provide further insights on how cellular processes such as innate immune sensing pathways, anti-apoptosis versus apoptosis may be differentially modulated in response to the virulent or vaccine strain of B. anthracis. The reported differences may account for the marked difference in pathogenicity between these two strains.

Cofactors required for TLR7- and TLR9-dependent innate immune responses.

Pathogens commonly utilize endocytic pathways to gain cellular access. The endosomal pattern recognition receptors TLR7 and TLR9 detect pathogen-encoded nucleic acids to initiate MyD88-dependent proinflammatory responses to microbial infection. Using genome-wide RNAi screening and integrative systems-based analysis, we identify 190 cofactors required for TLR7- and TLR9-directed signaling responses. A set of cofactors were crossprofiled for their activities downstream of several immunoreceptors and then functionally mapped based on the known architecture of NF-κB signaling pathways. Protein complexes and pathways involved in ubiquitin-protein ligase activities, sphingolipid metabolism, chromatin modifications, and ancient stress responses were found to modulate innate recognition of endosomal nucleic acids. Additionally, hepatocyte growth factor-regulated tyrosine kinase substrate (HRS) was characterized as necessary for ubiquitin-dependent TLR9 targeting to the endolysosome. Proteins and pathways identified here should prove useful in delineating strategies to manipulate innate responses for treatment of autoimmune disorders and microbial infection.

Chemical and biological properties of frequent screening hits.

High-throughput screening (HTS) has become an important technology for the drug discovery process. It has been noted that certain compounds frequently appear as hits in many screening campaigns. By mining an HTS database covering large chemical space and diverse biological functions, we identified many novel chemical features, as well as several biological processes that were associated with a significant portion of frequent hits. However, we also noted that several marketed drugs also contained characteristics that commonly were associated with frequent hits. This observation suggested that current generally employed strategies for triaging compounds may result in the removal of compounds with desirable properties. Therefore, we developed a novel strategy that overlaid chemical scaffolds and biological processes, along with empirical hit frequency data, in order to provide a more functional frequent hit triage strategy; the risk of removing biologically relevant frequent hits was reduced compared to the typical empirical hit frequency-based filtering strategy.

Imaging of Plasmodium liver stages to drive next-generation antimalarial drug discovery.

Most malaria drug development focuses on parasite stages detected in red blood cells, even though, to achieve eradication, next-generation drugs active against both erythrocytic and exo-erythrocytic forms would be preferable. We applied a multifactorial approach to a set of >4000 commercially available compounds with previously demonstrated blood-stage activity (median inhibitory concentration < 1 micromolar) and identified chemical scaffolds with potent activity against both forms. From this screen, we identified an imidazolopiperazine scaffold series that was highly enriched among compounds active against Plasmodium liver stages. The orally bioavailable lead imidazolopiperazine confers complete causal prophylactic protection (15 milligrams/kilogram) in rodent models of malaria and shows potent in vivo blood-stage therapeutic activity. The open-source chemical tools resulting from our effort provide starting points for future drug discovery programs, as well as opportunities for researchers to investigate the biology of exo-erythrocytic forms.