Alex Zhavoronkov

Applications of Deep Learning in Biomedicine

Increases in throughput and installed base of biomedical research equipment led to a massive accumulation of -omics data known to be highly variable, high-dimensional, and sourced from multiple often incompatible data platforms. While this data may be useful for biomarker identification and drug discovery, the bulk of it remains underutilized. Deep neural networks (DNNs) are efficient algorithms based on the use of compositional layers of neurons, with advantages well matched to the challenges -omics data presents. While achieving state-of-the-art results and even surpassing human accuracy in many challenging tasks, the adoption of deep learning in biomedicine has been comparatively slow. Here, we discuss key features of deep learning that may give this approach an edge over other machine learning methods. We then consider limitations and review a number of applications of deep learning in biomedical studies demonstrating proof of concept and practical utility.

Gadd45 proteins: relevance to aging, longevity and age-related pathologies.

The Gadd45 proteins have been intensively studied, in view of their important role in key cellular processes. Indeed, the Gadd45 proteins stand at the crossroad of the cell fates by controlling the balance between cell (DNA) repair, eliminating (apoptosis) or preventing the expansion of potentially dangerous cells (cell cycle arrest, cellular senescence), and maintaining the stem cell pool. However, the biogerontological aspects have not thus far received sufficient attention. Here we analyzed the pathways and modes of action by which Gadd45 members are involved in aging, longevity and age-related diseases. Because of their pleiotropic action, a decreased inducibility of Gadd45 members may have far-reaching consequences including genome instability, accumulation of DNA damage, and disorders in cellular homeostasis - all of which may eventually contribute to the aging process and age-related disorders (promotion of tumorigenesis, immune disorders, insulin resistance and reduced responsiveness to stress). Most recently, the dGadd45 gene has been identified as a longevity regulator in Drosophila. Although further wide-scale research is warranted, it is becoming increasingly clear that Gadd45s are highly relevant to aging, age-related diseases (ARDs) and to the control of life span, suggesting them as potential therapeutic targets in ARDs and pro-longevity interventions.

Brain-Computer Interface Based on Generation of Visual Images

This paper examines the task of recognizing EEG patterns that correspond to performing three mental tasks: relaxation and imagining of two types of pictures: faces and houses. The experiments were performed using two EEG headsets: BrainProducts ActiCap and Emotiv EPOC. The Emotiv headset becomes widely used in consumer BCI application allowing for conducting large-scale EEG experiments in the future. Since classification accuracy significantly exceeded the level of random classification during the first three days of the experiment with EPOC headset, a control experiment was performed on the fourth day using ActiCap. The control experiment has shown that utilization of high-quality research equipment can enhance classification accuracy (up to 68% in some subjects) and that the accuracy is independent of the presence of EEG artifacts related to blinking and eye movement. This study also shows that computationally-inexpensive Bayesian classifier based on covariance matrix analysis yields similar classification accuracy in this problem as a more sophisticated Multi-class Common Spatial Patterns (MCSP) classifier.

Genetics and epigenetics of aging and longevity

Evolutionary theories of aging predict the existence of certain genes that provide selective advantage early in life with adverse effect on lifespan later in life (antagonistic pleiotropy theory) or longevity insurance genes (disposable soma theory). Indeed, the study of human and animal genetics is gradually identifying new genes that increase lifespan when overexpressed or mutated: gerontogenes. Furthermore, genetic and epigenetic mechanisms are being identified that have a positive effect on longevity. The gerontogenes are classified as lifespan regulators, mediators, effectors, housekeeping genes, genes involved in mitochondrial function, and genes regulating cellular senescence and apoptosis. In this review we demonstrate that the majority of the genes as well as genetic and epigenetic mechanisms that are involved in regulation of longevity are highly interconnected and related to stress response.

Oncofinder, a new method for the analysis of intracellular signaling pathway activation using transcriptomic data

We propose a new biomathematical method, OncoFinder, for both quantitative and qualitative analysis of the intracellular signaling pathway activation (SPA). This method is universal and may be used for the analysis of any physiological, stress, malignancy and other perturbed conditions at the molecular level. In contrast to the other existing techniques for aggregation and generalization of the gene expression data for individual samples, we suggest to distinguish the positive/activator and negative/repressor role of every gene product in each pathway. We show that the relative importance of each gene product in a pathway can be assessed using kinetic models for “low-level” protein interactions. Although the importance factors for the pathway members cannot be so far established for most of the signaling pathways due to the lack of the required experimental data, we showed that ignoring these factors can be sometimes acceptable and that the simplified formula for SPA evaluation may be applied for many cases. We hope that due to its universal applicability, the method OncoFinder will be widely used by the researcher community.

Deep Learning Applications for Predicting Pharmacological Properties of Drugs and Drug Repurposing Using Transcriptomic Data

Deep learning is rapidly advancing many areas of science and technology with multiple success stories in image, text, voice and video recognition, robotics, and autonomous driving. In this paper we demonstrate how deep neural networks (DNN) trained on large transcriptional response data sets can classify various drugs to therapeutic categories solely based on their transcriptional profiles. We used the perturbation samples of 678 drugs across A549, MCF-7, and PC-3 cell lines from the LINCS Project and linked those to 12 therapeutic use categories derived from MeSH. To train the DNN, we utilized both gene level transcriptomic data and transcriptomic data processed using a pathway activation scoring algorithm, for a pooled data set of samples perturbed with different concentrations of the drug for 6 and 24 hours. In both pathway and gene level classification, DNN achieved high classification accuracy and convincingly outperformed the support vector machine (SVM) model on every multiclass classification problem, however, models based on pathway level data performed significantly better. For the first time we demonstrate a deep learning neural net trained on transcriptomic data to recognize pharmacological properties of multiple drugs across different biological systems and conditions. We also propose using deep neural net confusion matrices for drug repositioning. This work is a proof of principle for applying deep learning to drug discovery and development.

Molecular aspects of development and regulation of endometriosis.

Endometriosis is a common and painful condition affecting women of reproductive age. While the underlying pathophysiology is still largely unknown, much advancement has been made in understanding the progression of the disease. In recent years, a great deal of research has focused on non-invasive diagnostic tools, such as biomarkers, as well as identification of potential therapeutic targets. In this article, we will review the etiology and cellular mechanisms associated with endometriosis as well as the current diagnostic tools and therapies. We will then discuss the more recent genomic and proteomic studies and how these data may guide development of novel diagnostics and therapeutics. The current diagnostic tools are invasive and current therapies primarily treat the symptoms of endometriosis. Optimally, the advancement of “-omic” data will facilitate the development of non-invasive diagnostic biomarkers as well as therapeutics that target the pathophysiology of the disease and halt, or even reverse, progression. However, the amount of data generated by these types of studies is vast and bioinformatics analysis, such as we present here, will be critical to identification of appropriate targets for further study.

Human-specific endogenous retroviral insert serves as an enhancer for the schizophrenia-linked gene PRODH

Significance We identified a human-specific endogenous retroviral insert (hsERV) that acts as an enhancer for human PRODH, hsERV_PRODH. PRODH encodes proline dehydrogenase, which is involved in neuromediator synthesis in the CNS. We show that the hsERV_PRODH enhancer acts synergistically with the CpG island of PRODH and is regulated by methylation. We detected high PRODH expression in the hippocampus, which was correlated with the undermethylated state of this enhancer. PRODH regulatory elements provide neuron-specific transcription in hippocampal cells, and the mechanism of hsERV_PRODH enhancer activity involves the binding of transcriptional factor SOX2. Because PRODH is associated with several neurological disorders, we hypothesize that the human-specific regulation of PRODH by hsERV_PRODH may have played a role in human evolution by upregulating the expression of this important CNS-specific gene. Using a systematic, whole-genome analysis of enhancer activity of human-specific endogenous retroviral inserts (hsERVs), we identified an element, hsERVPRODH, that acts as a tissue-specific enhancer for the PRODH gene, which is required for proper CNS functioning. PRODH is one of the candidate genes for susceptibility to schizophrenia and other neurological disorders. It codes for a proline dehydrogenase enzyme, which catalyses the first step of proline catabolism and most likely is involved in neuromediator synthesis in the CNS. We investigated the mechanisms that regulate hsERVPRODH enhancer activity. We showed that the hsERVPRODH enhancer and the internal CpG island of PRODH synergistically activate its promoter. The enhancer activity of hsERVPRODH is regulated by methylation, and in an undermethylated state it can up-regulate PRODH expression in the hippocampus. The mechanism of hsERVPRODH enhancer activity involves the binding of the transcription factor SOX2, whch is preferentially expressed in hippocampus. We propose that the interaction of hsERVPRODH and PRODH may have contributed to human CNS evolution.

The OncoFinder algorithm for minimizing the errors introduced by the high-throughput methods of transcriptome analysis.

The diversity of the installed sequencing and microarray equipment make it increasingly difficult to compare and analyze the gene expression datasets obtained using the different methods. Many applications requiring high-quality and low error rates cannot make use of available data using traditional analytical approaches. Recently, we proposed a new concept of signalome-wide analysis of functional changes in the intracellular pathways termed OncoFinder, a bioinformatic tool for quantitative estimation of the signaling pathway activation (SPA). We also developed methods to compare the gene expression data obtained using multiple platforms and minimizing the error rates by mapping the gene expression data onto the known and custom signaling pathways. This technique for the first time makes it possible to analyze the functional features of intracellular regulation on a mathematical basis. In this study we show that the OncoFinder method significantly reduces the errors introduced by transcriptome-wide experimental techniques. We compared the gene expression data for the same biological samples obtained by both the next generation sequencing (NGS) and microarray methods. For these different techniques we demonstrate that there is virtually no correlation between the gene expression values for all datasets analyzed (R2 < 0.1). In contrast, when the OncoFinder algorithm is applied to the data we observed clear-cut correlations between the NGS and microarray gene expression datasets. The SPA profiles obtained using NGS and microarray techniques were almost identical for the same biological samples allowing for the platform-agnostic analytical applications. We conclude that this feature of the OncoFinder enables to characterize the functional states of the transcriptomes and interactomes more accurately as before, which makes OncoFinder a method of choice for many applications including genetics, physiology, biomedicine, and molecular diagnostics.

Characteristic patterns of microrna expression in human bladder cancer

MicroRNAs (miRNAs) are small, non-coding RNAs that post-transcriptionally regulate gene expression. Their altered expression and functional activity have been observed in many human cancers. miRNAs represent promising diagnostic and prognostic molecular biomarkers, and also serve as novel therapeutic targets. We performed a systematic analysis of scientific reports that link differences in miRNA expression with the pathogenesis of bladder cancer (BC). This literature review is the first comprehensive database of miRNA molecules with biased expression profiles in BC. Among the 95 differentially expressed miRNAs that we identified from the literature, we classify 48 as up-regulated in BC, 35 as down-regulated, and 12 as contradictory (contradictory data were reported in one or more studies on the gene). In addition, we discuss the possible roles of differentially expressed miRNAs in the regulation of intracellular signaling pathways in BC.