Industry-scale application and evaluation of deep learning for drug target prediction.

Artificial intelligence (AI) is undergoing a revolution thanks to the breakthroughs of machine learning algorithms in computer vision, speech recognition, natural language processing and generative modelling. Recent works on publicly available pharmaceutical data showed that AI methods are highly promising for Drug Target prediction. However, the quality of public data might be different than that of industry data due to different labs reporting measurements, different measurement techniques, fewer samples and less diverse and specialized assays. As part of a European funded project (ExCAPE), that brought together expertise from pharmaceutical industry, machine learning, and high-performance computing, we investigated how well machine learning models obtained from public data can be transferred to internal pharmaceutical industry data. Our results show that machine learning models trained on public data can indeed maintain their predictive power to a large degree when applied to industry data. Moreover, we observed that deep learning derived machine learning models outperformed comparable models, which were trained by other machine learning algorithms, when applied to internal pharmaceutical company datasets. To our knowledge, this is the first large-scale study evaluating the potential of machine learning and especially deep learning directly at the level of industry-scale settings and moreover investigating the transferability of publicly learned target prediction models towards industrial bioactivity prediction pipelines.

Retinoic acid receptor α as a novel contributor to adrenal cortex structure and function through interactions with Wnt and Vegfa signalling.

Primary aldosteronism (PA) is the most frequent form of secondary arterial hypertension. Mutations in different genes increase aldosterone production in PA, but additional mechanisms may contribute to increased cell proliferation and aldosterone producing adenoma (APA) development. We performed transcriptome analysis in APA and identified retinoic acid receptor alpha (RARα) signaling as a central molecular network involved in nodule formation. To understand how RARα modulates adrenal structure and function, we explored the adrenal phenotype of male and female Rarα knockout mice. Inactivation of Rarα in mice led to significant structural disorganization of the adrenal cortex in both sexes, with increased adrenal cortex size in female mice and increased cell proliferation in males. Abnormalities of vessel architecture and extracellular matrix were due to decreased Vegfa expression and modifications in extracellular matrix components. On the molecular level, Rarα inactivation leads to inhibition of non-canonical Wnt signaling, without affecting the canonical Wnt pathway nor PKA signaling. Our study suggests that Rarα contributes to the maintenance of normal adrenal cortex structure and cell proliferation, by modulating Wnt signaling. Dysregulation of this interaction may contribute to abnormal cell proliferation, creating a propitious environment for the emergence of specific driver mutations in PA.

ExCAPE-DB: an integrated large scale dataset facilitating Big Data analysis in chemogenomics.

Chemogenomics data generally refers to the activity data of chemical compounds on an array of protein targets and represents an important source of information for building in silico target prediction models. The increasing volume of chemogenomics data offers exciting opportunities to build models based on Big Data. Preparing a high quality data set is a vital step in realizing this goal and this work aims to compile such a comprehensive chemogenomics dataset. This dataset comprises over 70 million SAR data points from publicly available databases (PubChem and ChEMBL) including structure, target information and activity annotations. Our aspiration is to create a useful chemogenomics resource reflecting industry-scale data not only for building predictive models of in silico polypharmacology and off-target effects but also for the validation of cheminformatics approaches in general.

KCNJ5 mutations in aldosterone producing adenoma and relationship with adrenal cortex remodeling.

Somatic mutations of KCNJ5, coding for the potassium channel GIRK4, have recently been implicated in the formation of aldosterone producing adenoma (APA). While a causal link between KCNJ5 mutations, membrane depolarization and aldosterone production has been established, the precise mechanism by which these mutations promote cell proliferation and APA formation remains unclear. The aim of our study was to correlate KCNJ5 mutation status with morphological and functional characteristics of the adrenal cortex adjacent to APA. While GIRK4 was expressed in APA and in the zona glomerulosa of the adjacent cortex, significantly lower levels were detected in APA harboring a KCNJ5 mutation. There was no correlation between KCNJ5 mutation status and the morphological measures of adrenal cortex remodeling, including nodulation, vascularization and expression of CYP11B2. The cell composition of APA was not significantly different between groups. These results indicate that KCNJ5 mutations are not correlated with adrenal cortex remodeling in APA.

A network perspective on metabolic inconsistency.

BACKGROUND: Integrating gene expression profiles and metabolic pathways under different experimental conditions is essential for understanding the coherence of these two layers of cellular organization. The network character of metabolic systems can be instrumental in developing concepts of agreement between expression data and pathways. A network-driven interpretation of gene expression data has the potential of suggesting novel classifiers for pathological cellular states and of contributing to a general theoretical understanding of gene regulation. RESULTS: Here, we analyze the coherence of gene expression patterns and a reconstruction of human metabolism, using consistency scores obtained from network and constraint-based analysis methods. We find a surprisingly strong correlation between the two measures, demonstrating that a substantial part of inconsistencies between metabolic processes and gene expression can be understood from a network perspective alone. Prompted by this finding, we investigate the topological context of the individual biochemical reactions responsible for the observed inconsistencies. On this basis, we are able to separate the differential contributions that bear physiological information about the system, from the unspecific contributions that unravel gaps in the metabolic reconstruction. We demonstrate the biological potential of our network-driven approach by analyzing transcriptome profiles of aldosterone producing adenomas that have been obtained from a cohort of Primary Aldosteronism patients. We unravel systematics in the data that could not have been resolved by conventional microarray data analysis. In particular, we discover two distinct metabolic states in the adenoma expression patterns. CONCLUSIONS: The methodology presented here can help understand metabolic inconsistencies from a network perspective. It thus serves as a mediator between the topology of metabolic systems and their dynamical function. Finally, we demonstrate how physiologically relevant insights into the structure and dynamics of metabolic networks can be obtained using this novel approach.

Prevalence, clinical, and molecular correlates of KCNJ5 mutations in primary aldosteronism.

Primary aldosteronism is the most common form of secondary hypertension. Mutations in the KCNJ5 gene have been described recently in aldosterone-producing adenomas (APAs). The aim of this study was to investigate the prevalence of KCNJ5 mutations in unselected patients with primary aldosteronism and their clinical, biological and molecular correlates. KCNJ5 sequencing was performed on somatic (APA, n=380) and peripheral (APA, n=344; bilateral adrenal hyperplasia, n=174) DNA of patients with primary aldosteronism, collected through the European Network for the Study of Adrenal Tumors. Transcriptome analysis was performed in 102 tumors. Somatic KCNJ5 mutations (p.Gly151Arg or p.Leu168Arg) were found in 34% (129 of 380) of APA. They were significantly more prevalent in females (49%) than males (19%; P<10(-3)) and in younger patients (42.1±1.0 versus 47.6±0.7 years; P<10(-3)) and were associated with higher preoperative aldosterone levels (455±26 versus 376±17 ng/L; P=0.012) but not with therapeutic outcome after surgery. Germline KCNJ5 mutations were found neither in patients with APA nor those with bilateral adrenal hyperplasia. Somatic KCNJ5 mutations were specific for APA, because they were not identified in 25 peritumoral adrenal tissues or 16 cortisol-producing adenomas. Hierarchical clustering of transcriptome profiles showed that APAs with p.Gly151Arg or p.Leu168Arg mutations were indistinguishable from tumors without KCNJ5 mutations. In conclusion, although a large proportion of sporadic APAs harbors somatic KCNJ5 mutations, germline mutations are not similarly causative for bilateral adrenal hyperplasia. KCNJ5 mutation carriers are more likely to be females; younger age and higher aldosterone levels at diagnosis suggest that KCNJ5 mutations may be associated with a more florid phenotype of primary aldosteronism.

Aldosterone-producing adenoma formation in the adrenal cortex involves expression of stem/progenitor cell markers.

Aldosterone producing adenoma (APA) is the most common form of surgically curable hypertension. To further understand mechanisms involved in APA formation, we investigated the expression of molecules linked to adrenal stem/precursor cells [ß-catenin, Sonic hedgehog (Shh), CD56], and nuclear receptors that play key roles in adrenocortical development and function steroidogenic factor 1, dosage-sensitive sex reversal-adrenal hypoplasia congenita critical region on the X chromosome, gene 1) in six control adrenal glands and 14 adrenals with APA and compared their expression with that of specific markers of zona glomerulosa (ZG) [CYP11B2, Disabled 2 (Dab2)]. Both Dab2 and CD56 were expressed in ZG. Although Dab2 associates uniquely with differentiated ZG cells and its expression is lost when cells transdifferentiate to zona fasciculata (ZF) cells, CD56 was also expressed in ZF and in aldosterone-producing cell clusters, confirming that these structures possess an intermediate phenotype between ZG and ZF cells. Shh was barely detectable in cells located to the outer part of the ZG in the control adrenal; in contrast, its expression was detected in the entire APA and was dramatically increased in the hyperplastic peritumoral ZG. Transcriptome profiling revealed differential expression of components of Shh signaling pathway in a subgroup of APA. Similarly, Wnt/ß-catenin signaling was activated in the majority of APA as well as in the entire peritumoral adrenal cortex; however, no mutation was identified in the CTNNB1 gene that could account for ß-catenin activation. Our data suggest that both APA and adjacent ZG present characteristics of stem/precursor cells; the reexpression of genes involved in fetal adrenal development could underlie excessive ZG cell proliferation and APA formation.

Glucocorticoid receptors in dopaminoceptive neurons, key for cocaine, are dispensable for molecular and behavioral morphine responses.

BACKGROUND: Psychostimulants and opiates trigger similar enduring neuroadaptations within the reward circuitry thought to underlie addiction. Transcription factors are key to mediating these enduring behavioral alterations. The facilitation of these maladaptive changes by glucocorticoid hormones suggests that the glucocorticoid receptor (GR), a transcription factor involved in the stress response, could be a common mediator of responses to pharmacologically distinct classes of abused drugs. METHODS: We employed mouse models carrying GR gene inactivation in either dopamine or dopaminoceptive neurons to determine the involvement of this transcription factor in behavioral responses to cocaine and morphine. We then combined microarray analysis, drug-elicited immediate early gene induction, and in vivo microdialysis to elucidate the molecular underpinnings of these responses. RESULTS: Inactivating GR within dopaminoceptive neurons markedly reduces cocaine-induced conditioned place preference and the expression of locomotor sensitization. In striking contrast, GR had no effect on behavioral morphine responses in either dopaminoceptive or dopamine neurons. The dopaminoceptive mutation engenders alterations in the expression of striatal genes that are implicated in glutamatergic transmission and plasticity. Within the nucleus accumbens, impaired cellular responses to cocaine are conspicuous; a pronounced deficit in cocaine-elicited extracellular dopamine release, expression of the key IEGs c-Fos and Zif268, and phosphorylation of extracellular signal-regulated kinases 1/2 in mutants were observed. In contrast, these molecular and neurochemical changes were not observed in response to morphine, mirroring the lack of effect on behavioral responses to morphine. CONCLUSION: Combined behavioral and molecular approaches have identified a subset of neurons in which GR differentially influences cocaine- and morphine-induced responses.