Predicting clinically promising therapeutic hypotheses using tensor factorization.

BACKGROUND: Determining which target to pursue is a challenging and error-prone first step in developing a therapeutic treatment for a disease, where missteps are potentially very costly given the long-time frames and high expenses of drug development. With current informatics technology and machine learning algorithms, it is now possible to computationally discover therapeutic hypotheses by predicting clinically promising drug targets based on the evidence associating drug targets with disease indications. We have collected this evidence from Open Targets and additional databases that covers 17 sources of evidence for target-indication association and represented the data as a tensor of 21,437 × 2211 × 17. RESULTS: As a proof-of-concept, we identified examples of successes and failures of target-indication pairs in clinical trials across 875 targets and 574 disease indications to build a gold-standard data set of 6140 known clinical outcomes. We designed and executed three benchmarking strategies to examine the performance of multiple machine learning models: Logistic Regression, LASSO, Random Forest, Tensor Factorization and Gradient Boosting Machine. With 10-fold cross-validation, tensor factorization achieved AUROC = 0.82 ± 0.02 and AUPRC = 0.71 ± 0.03. Across multiple validation schemes, this was comparable or better than other methods. CONCLUSION: In this work, we benchmarked a machine learning technique called tensor factorization for the problem of predicting clinical outcomes of therapeutic hypotheses. Results have shown that this method can achieve equal or better prediction performance compared with a variety of baseline models. We demonstrate one application of the method to predict outcomes of trials on novel indications of approved drug targets. This work can be expanded to targets and indications that have never been clinically tested and proposing novel target-indication hypotheses. Our proposed biologically-motivated cross-validation schemes provide insight into the robustness of the prediction performance. This has significant implications for all future methods that try to address this seminal problem in drug discovery.

Systematic interrogation of diverse Omic data reveals interpretable, robust, and generalizable transcriptomic features of clinically successful therapeutic targets.

Target selection is the first and pivotal step in drug discovery. An incorrect choice may not manifest itself for many years after hundreds of millions of research dollars have been spent. We collected a set of 332 targets that succeeded or failed in phase III clinical trials, and explored whether Omic features describing the target genes could predict clinical success. We obtained features from the recently published comprehensive resource: Harmonizome. Nineteen features appeared to be significantly correlated with phase III clinical trial outcomes, but only 4 passed validation schemes that used bootstrapping or modified permutation tests to assess feature robustness and generalizability while accounting for target class selection bias. We also used classifiers to perform multivariate feature selection and found that classifiers with a single feature performed as well in cross-validation as classifiers with more features (AUROC = 0.57 and AUPR = 0.81). The two predominantly selected features were mean mRNA expression across tissues and standard deviation of expression across tissues, where successful targets tended to have lower mean expression and higher expression variance than failed targets. This finding supports the conventional wisdom that it is favorable for a target to be present in the tissue(s) affected by a disease and absent from other tissues. Overall, our results suggest that it is feasible to construct a model integrating interpretable target features to inform target selection. We anticipate deeper insights and better models in the future, as researchers can reuse the data we have provided to improve methods for handling sample biases and learn more informative features. Code, documentation, and data for this study have been deposited on GitHub at https://github.com/arouillard/omic-features-successful-targets.

Inhibition of p38 mitogen-activated protein kinase reduces inflammation after coronary vascular injury in humans.

OBJECTIVE: To evaluate whether a p38α/ß mitogen-activated protein kinase inhibitor, SB-681323, would limit the elevation of an inflammatory marker, high-sensitivity C-reactive protein (hsCRP), after a percutaneous coronary intervention (PCI). METHODS AND RESULTS: Coronary artery stents provide benefit by maintaining lumen patency but may incur vascular trauma and inflammation, leading to myocardial damage. A key mediator for such stress signaling is p38 mitogen-activated protein kinase. Patients with angiographically documented coronary artery disease receiving stable statin therapy and about to undergo PCI were randomly selected to receive SB-681323, 7.5 mg (n=46), or placebo (n=46) daily for 28 days, starting 3 days before PCI. On day 3, before PCI, hsCRP was decreased in the SB-681323 group relative to the placebo group (29% lower; P=0.02). After PCI, there was a statistically significant attenuation in the increase in hsCRP in the SB-681323 group relative to the placebo group (37% lower on day 5 [P=0.04]; and 40% lower on day 28 [P=0.003]). There were no adverse safety signals after 28 days of treatment with SB-681323. CONCLUSIONS: In the setting of statin therapy, SB-681323 significantly attenuated the post-PCI inflammatory response, as measured by hsCRP. This inflammatory dampening implicates p38 mitogen-activated protein kinase in the poststent response, potentially defining an avenue to limit poststent restenosis.

Network and pathway expansion of genetic disease associations identifies successful drug targets.

Genetic evidence of disease association has often been used as a basis for selecting of drug targets for complex common diseases. Likewise, the propagation of genetic evidence through gene or protein interaction networks has been shown to accurately infer novel disease associations at genes for which no direct genetic evidence can be observed. However, an empirical test of the utility of combining these approaches for drug discovery has been lacking. In this study, we examine genetic associations arising from an analysis of 648 UK Biobank GWAS and evaluate whether targets identified as proxies of direct genetic hits are enriched for successful drug targets, as measured by historical clinical trial data. We find that protein networks formed from specific functional linkages such as protein complexes and ligand-receptor pairs are suitable for even naïve guilt-by-association network propagation approaches. In addition, more sophisticated approaches applied to global protein-protein interaction networks and pathway databases, also successfully retrieve targets enriched for clinically successful drug targets. We conclude that network propagation of genetic evidence can be used for drug target identification.

Phenome-wide Mendelian randomization mapping the influence of the plasma proteome on complex diseases.

The human proteome is a major source of therapeutic targets. Recent genetic association analyses of the plasma proteome enable systematic evaluation of the causal consequences of variation in plasma protein levels. Here we estimated the effects of 1,002 proteins on 225 phenotypes using two-sample Mendelian randomization (MR) and colocalization. Of 413 associations supported by evidence from MR, 130 (31.5%) were not supported by results of colocalization analyses, suggesting that genetic confounding due to linkage disequilibrium is widespread in naïve phenome-wide association studies of proteins. Combining MR and colocalization evidence in cis-only analyses, we identified 111 putatively causal effects between 65 proteins and 52 disease-related phenotypes ( https://www.epigraphdb.org/pqtl/ ). Evaluation of data from historic drug development programs showed that target-indication pairs with MR and colocalization support were more likely to be approved, evidencing the value of this approach in identifying and prioritizing potential therapeutic targets.

Pathway analysis of GWAS loci identifies novel drug targets and repurposing opportunities.

Genome-wide association studies (GWAS) have made considerable progress and there is emerging evidence that genetics-based targets can lead to 28% more launched drugs. We analyzed 1589 GWAS across 1456 pathways to translate these often imprecise genetic loci into therapeutic hypotheses for 182 diseases. These pathway-based genetic targets were validated by testing whether current drug targets were enriched in the pathway space for the same indication. Remarkably, 30% of diseases had significantly more targets in these pathways than expected by chance; the comparable number for GWAS alone (without pathway analysis) was zero. This study shows that a systematic global pathway analysis can translate genetic findings into therapeutic hypotheses for both new drug discovery and repositioning opportunities for current drugs.

Preclinical evaluation of EPHX2 inhibition as a novel treatment for inflammatory bowel disease.

Epoxyeicosatrienoic acids (EETs) are signaling lipids produced by cytochrome P450 epoxygenation of arachidonic acid, which are metabolized by EPHX2 (epoxide hydrolase 2, alias soluble epoxide hydrolase or sEH). EETs have pleiotropic effects, including anti-inflammatory activity. Using a Connectivity Map (CMAP) approach, we identified an inverse-correlation between an exemplar EPHX2 inhibitor (EPHX2i) compound response and an inflammatory bowel disease patient-derived signature. To validate the gene-disease link, we tested a pre-clinical tool EPHX2i (GSK1910364) in a mouse disease model, where it showed improved outcomes comparable to or better than the positive control Cyclosporin A. Up-regulation of cytoprotective genes and down-regulation of proinflammatory cytokine production were observed in colon samples obtained from EPHX2i-treated mice. Follow-up immunohistochemistry analysis verified the presence of EPHX2 protein in infiltrated immune cells from Crohn's patient tissue biopsies. We further demonstrated that GSK2256294, a clinical EPHX2i, reduced the production of IL2, IL12p70, IL10 and TNFα in both ulcerative colitis and Crohn's disease patient-derived explant cultures. Interestingly, GSK2256294 reduced IL4 and IFNγ in ulcerative colitis, and IL1ß in Crohn's disease specifically, suggesting potential differential effects of GSK2256294 in these two diseases. Taken together, these findings suggest a novel therapeutic use of EPHX2 inhibition for IBD.

Predominance of a proinflammatory phenotype in monocyte-derived macrophages from subjects with low plasma HDL-cholesterol.

OBJECTIVE: Reduced plasma concentrations of high-density lipoprotein-cholesterol (HDL-C) are a significant risk factor for cardiovascular disease. Mechanisms that regulate HDL-C concentrations represent an important area of investigation. METHODS AND RESULTS: Comparative transcriptome analyses of monocyte-derived macrophages (MDM) from a large population of low HDL-C subjects and age- and sex-matched controls revealed a cluster of inflammatory genes highly expressed in low HDL-C subjects. The expression levels of peroxisome proliferator activated receptor (PPAR) gamma and several antioxidant metallothionein genes were decreased in MDM from all low HDL-C groups compared with controls, as was the expression of other genes regulated by PPARgamma, including CD36, adipocyte fatty acid binding protein (FABP4), and adipophilin (ADFP). In contrast, PPARdelta expression was increased in MDM from low HDL-C groups. Quantitative RT-PCR corroborated all major findings from the microarray analysis in two separate patient cohorts. Expression of several inflammatory cytokine genes including interleukin 1beta, interleukin 8, and tumor necrosis factor alpha were highly increased in low HDL-C subjects. CONCLUSIONS: The activated proinflammatory state of monocytes and MDM in low HDL-C subjects constitutes a novel parameter of risk associated with HDL deficiency, related to altered expression of metallothionein genes and the reciprocal regulation of PPARgamma and PPARdelta.

Chfr regulates a mitotic stress pathway through its RING-finger domain with ubiquitin ligase activity.

Resistance to chemotherapy targeting microtubules could be partially because of the delay in chromosome condensation and segregation during mitosis. The Chfr pathway has been defined recently, and its activation causes a delay in chromosome condensation in response to mitotic stress. Because Chfr contains a RING-finger domain, we tested whether Chfr inhibits chromosome condensation through an ubiquitin (ubiquitin)-dependent pathway. In the presence of purified E1, Ubc4, or Ubc5, and ubiquitin, Chfr catalyzes its own ubiquitination in vitro, an activity requiring the RING domain. In vivo, overexpressed Chfr but not a RING domain mutant is spontaneously ubiquitinated. Our studies with DLD1 cells stably expressing wild-type Chfr and Chfr lacking the RING domain indicated that the RING-finger deletion mutant was defective in inhibiting chromosome condensation after Taxol treatment. In addition, Chfr expression increases the survival rate after Taxol treatment, an activity requiring the RING domain. Preliminary studies indicate that Chfr expression is cell cycle regulated and is dependent on its ubiquitin ligase activity. It is very likely that the Chfr-mediated ubiquitin-dependent pathway is a critical component of the response to mitotic stress.

Systematic Analysis of Drug Targets Confirms Expression in Disease-Relevant Tissues.

It is commonly assumed that drug targets are expressed in tissues relevant to their indicated diseases, even under normal conditions. While multiple anecdotal cases support this hypothesis, a comprehensive study has not been performed to verify it. We conducted a systematic analysis to assess gene and protein expression for all targets of marketed and phase III drugs across a diverse collection of normal human tissues. For 87% of gene-disease pairs, the target is expressed in a disease-affected tissue under healthy conditions. This result validates the importance of confirming expression of a novel drug target in an appropriate tissue for each disease indication and strengthens previous findings showing that targets of efficacious drugs should be expressed in relevant tissues under normal conditions. Further characterization of the remaining 13% of gene-disease pairs revealed that most genes are expressed in a different tissue linked to another disease. Our analysis demonstrates the value of extensive tissue specific expression resources.both in terms of tissue and cell diversity as well as techniques used to measure gene expression.

Impact of genetically supported target selection on R&D productivity.

This corrects the article DOI: 10.1038/nrd.2016.164.

Revealing transforming growth factor-beta signaling transduction in human kidney by gene expression data mining.

Tumor growth factor-beta (TGF-beta) is a key mediator of glomerular and tubulointerstitial pathobiology in chronic kidney disease. Its signaling transduction controls a diverse number of biological processes in a dynamic and context-dependent manner. We applied a data mining strategy to deconvolute gene expression patterns across hundreds of microarray data sets to reveal members of the TGF-beta signaling network in human kidney. This strategy is composed of three major steps: (i) select genes known to be involved and expressionally regulated in TGF-beta signaling as "bait"; (ii) select microarray data sets in which the bait genes are strongly co-regulated; (iii) identify (or "fish") additional TGF-beta signaling genes by a non-parametric statistic-based gene scoring system (NP score). The 40 genes with highest NP scores and significant permutation p values were selected for in silico validation, and used to identify a network, in which 35 of these genes were found to be connected by literature- derived relationships. Transcription factors were found to be enriched in the top list. Among them, activated transcription factor 3 (ATF3) had the highest NP score, and was proposed to play a pivotal role in TGF-beta signaling in human kidney. Finally, we implemented a non-parametric pathway ranking (NPPR) tool (Mootha et al., 2003) to rank pathways and identified canonical biological pathways associated with the down-stream of TGF-beta signaling.

Correlation of protein and gene expression profiles of inflammatory proteins after endotoxin challenge in human subjects.

Administration of endotoxin (LPS) in humans results in profound physiological responses, including activation of peripheral blood mononuclear cells and the release of inflammatory factors. The time course of the response of selected inflammatory proteins was examined in healthy subjects (n = 6) administered a single intravenous dose of the purified derivative of endotoxin (3.0 ng/kg). Microarray analysis demonstrated changes in the expression of a number of genes, which were confirmed in separate in vitro endotoxin stimulation experiments. Subsequent TaqMan analysis of genes of interest indicated time-dependent changes in the expression of many of these genes. This included pre-B cell enhancing factor, which was identified on microarray analysis as being markedly upregulated following endotoxin stimulation. Protein expression of the genes examined by TaqMan analysis was measured and demonstrated the appearance of tumor necrosis factor (TNF)-alpha and sTNF-R proteins in the plasma beginning within 1 h after dosing, followed by other cytokines/ inflammatory markers (e.g., IL-1ra, G-CSF, IL-6, IL-8, and IL-10) and suppressors of cytokine signaling (SOCS-1 and SOCS-3). In general, cytokine protein expression correlated well with gene expression; however, the temporal profile of expression of some genes did not correlate well with the protein data. For many of these proteins, the lack of correlation was attributable to alternate tissue sources, which were demonstrated on TaqMan analysis. Principal component analysis indicated that cytokines could be grouped according to their temporal pattern of response, with most transcript levels returning to baseline 24 h following endotoxin administration. The combination of cDNA microarray and TaqMan analysis to identify and quantify changes in gene expression, along with the analysis of protein expression, can be useful in investigating inflammatory and other diseases.

Release of 50 new, drug-like compounds and their computational target predictions for open source anti-tubercular drug discovery.

As a follow up to the antimycobacterial screening exercise and the release of GSK´s first Tres Cantos Antimycobacterial Set (TCAMS-TB), this paper presents the results of a second antitubercular screening effort of two hundred and fifty thousand compounds recently added to the GSK collection. The compounds were further prioritized based on not only antitubercular potency but also on physicochemical characteristics. The 50 most attractive compounds were then progressed for evaluation in three different predictive computational biology algorithms based on structural similarity or GSK historical biological assay data in order to determine their possible mechanisms of action. This effort has resulted in the identification of novel compounds and their hypothesized targets that will hopefully fuel future TB drug discovery and target validation programs alike.

Combined Analysis of Phenotypic and Target-Based Screening in Assay Networks.

Small-molecule screens are an integral part of drug discovery. Public domain data in PubChem alone represent more than 158 million measurements, 1.2 million molecules, and 4300 assays. We conducted a global analysis of these data, building a network of assays and connecting the assays if they shared nonpromiscuous active molecules. This network spans both phenotypic and target-based screens, recapitulates known biology, and identifies new polypharmacology. Phenotypic screens are extremely important for drug discovery, contributing to the discovery of a large proportion of new drugs. Connections between phenotypic and biochemical, target-based screens can suggest strategies for repurposing both small-molecule and biologic drugs. For example, a screen for molecules that prevent cell death from a mutated version of superoxide-dismutase is linked with ALOX15. This connection suggests a therapeutic role for ALOX15 inhibitors in amyotrophic lateral sclerosis. An interactive version of the network is available online (http://swami.wustl.edu/flow/assay_network.html).

Evaluation of analytical methods for connectivity map data.

Connectivity map data and associated methodologies have become a valuable tool in understanding drug mechanism of action (MOA) and discovering new indications for drugs. However, few systematic evaluations have been done to assess the accuracy of these methodologies. One of the difficulties has been the lack of benchmarking data sets. Iskar et al. (PLoS. Comput. Biol. 6, 2010) predicted the Anatomical Therapeutic Chemical (ATC) drug classification based on drug-induced gene expression profile similarity (DIPS), and quantified the accuracy of their method by computing the area under the curve (AUC) of the Receiver Operating Characteristic (ROC) curve. We adopt the same data and extend the methodology, by using a simpler eXtreme cosine (XCos) method, and find it does better in this limited setting than the Kolmogorov-Smirnov (KS) statistic. In fact, for partial AUC (a more relevant statistic for actual application to repositioning) XCos does 17% better than the DIPS method (p=1.2e-7). We also observe that smaller gene signatures (with 100 probes) do better than larger ones (with 500 probes), and that DMSO controls from within the same batch obviate the need for mean centering. As expected there is heterogeneity in the prediction accuracy amongst the various ATC codes. We find that good transcriptional response to drug treatment appears necessary but not sufficient to achieve high AUCs. Certain ATC codes, such as those corresponding to corticosteroids, had much higher AUCs possibly due to strong transcriptional responses and consistency in MOA.