Search

Surprisingly promising results have been achieved by deep learning (DL) systems in recent years. Many of these achievements have been reached in academic settings, or by large technology companies with highly skilled research groups and advanced supporting infrastructure. For companies without large research groups or advanced infrastructure, building high-quality production-ready systems with DL components has proven challenging. There is a clear lack of well-functioning tools and best practices for building DL systems. It is the goal of this research to identify what the main challenges are, by applying an interpretive research approach in close collaboration with companies of varying size and type. A set of seven projects have been selected to describe the potential with this new technology and to identify associated main challenges. A set of 12 main challenges has been identified and categorized into the three areas of development, production, and organizational challenges. Furthermore, a mapping between the challenges and the projects is defined, together with selected motivating descriptions of how and why the challenges apply to specific projects. Compared to other areas such as software engineering or database technologies, it is clear that DL is still rather immature and in need of further work to facilitate development of high-quality systems. The challenges identified in this paper can be used to guide future research by the software engineering and DL communities. Together, we could enable a large number of companies to start taking advantage of the high potential of the DL technology., Comment: 44th Euromicro Conference on Software Engineering and Advanced Applications, IEEE, 2018

Complex regulatory dynamics is ubiquitous in molecular networks composed of genes and proteins. Recent progress in computational biology and its application to molecular data generate a growing number of complex networks. Yet, it has been difficult to understand the governing principles of these networks beyond graphical analysis or extensive numerical simulations. Here the authors exploit several simplifying biological circumstances which thereby enable to directly detect the underlying dynamical regularities driving periodic oscillations in a dynamical nonlinear computational model of a protein-protein network. System analysis is performed using the cell cycle, a mathematically well-described complex regulatory circuit driven by external signals. By introducing an explicit time delay and using a -tearing-and-zooming- approach the authors reduce the system to a piecewise linear system with two variables that capture the dynamics of this complex network. A key step in the analysis is the identification of functional subsystems by identifying the relations between state-variables within the model. These functional subsystems are referred to as dynamical modules operating as sensitive switches in the original complex model. By using reduced mathematical representations of the subsystems the authors derive explicit conditions on how the cell cycle dynamics depends on system parameters, and can, for the first time, analyse and prove global conditions for system stability. The approach which includes utilising biological simplifying conditions, identification of dynamical modules and mathematical reduction of the model complexity may be applicable to other well-characterised biological regulatory circuits.

Environmental exposures filtered through the genetic make-up of each individual alter the transcriptional repertoire in organs central to metabolic homeostasis, thereby affecting arterial lipid accumulation, inflammation, and the development of coronary artery disease (CAD). The primary aim of the Stockholm Atherosclerosis Gene Expression (STAGE) study was to determine whether there are functionally associated genes (rather than individual genes) important for CAD development. To this end, two-way clustering was used on 278 transcriptional profiles of liver, skeletal muscle, and visceral fat (n =66/tissue) and atherosclerotic and unaffected arterial wall (n =40/tissue) isolated from CAD patients during coronary artery bypass surgery. The first step, across all mRNA signals (n =15,042/12,621 RefSeqs/genes) in each tissue, resulted in a total of 60 tissue clusters (n= 3958 genes). In the second step (performed within tissue clusters), one atherosclerotic lesion (n =49/48) and one visceral fat (n =59) cluster segregated the patients into two groups that differed in the extent of coronary stenosis (P=0.008 and P=0.00015). The associations of these clusters with coronary atherosclerosis were validated by analyzing carotid atherosclerosis expression profiles. Remarkably, in one cluster (n =55/54) relating to carotid stenosis (P =0.04), 27 genes in the two clusters relating to coronary stenosis were confirmed (n= 16/17, P<10 -27and-30). Genes in the transendothelial migration of leukocytes (TEML) pathway were overrepresented in all three clusters, referred to as the atherosclerosis module (A-module). In a second validation step, using three independent cohorts, the Amodule was found to be genetically enriched with CAD risk by 1.8-fold (P<0.004). The transcription co-factor LIM domain binding 2 (LDB2) was identified as a potential high-hierarchy regulator of the A-module, a notion supported by subnetwork analysis, by cellular and lesion expression of LDB2, and by the

Complex regulatory dynamics is ubiquitous in molecular networks composed of genes and proteins. Recent progress in computational biology and its application to molecular data generate a growing number of complex networks. Yet, it has been difficult to understand the governing principles of these networks beyond graphical analysis or extensive numerical simulations. Here the authors exploit several simplifying biological circumstances which thereby enable to directly detect the underlying dynamical regularities driving periodic oscillations in a dynamical nonlinear computational model of a protein-protein network. System analysis is performed using the cell cycle, a mathematically well-described complex regulatory circuit driven by external signals. By introducing an explicit time delay and using a tearing-and-zooming approach the authors reduce the system to a piecewise linear system with two variables that capture the dynamics of this complex network. A key step in the analysis is the identification of functional subsystems by identifying the relations between state-variables within the model. These functional subsystems are referred to as dynamical modules operating as sensitive switches in the original complex model. By using reduced mathematical representations of the subsystems the authors derive explicit conditions on how the cell cycle dynamics depends on system parameters, and can, for the first time, analyse and prove global conditions for system stability. The approach which includes utilising biological simplifying conditions, identification of dynamical modules and mathematical reduction of the model complexity may be applicable to other well-characterised biological regulatory circuits.

Environmental exposures filtered through the genetic make-up of each individual alter the transcriptional repertoire in organs central to metabolic homeostasis, thereby affecting arterial lipid accumulation, inflammation, and the development of coronary artery disease (CAD). The primary aim of the Stockholm Atherosclerosis Gene Expression (STAGE) study was to determine whether there are functionally associated genes (rather than individual genes) important for CAD development. To this end, two-way clustering was used on 278 transcriptional profiles of liver, skeletal muscle, and visceral fat (n=66/tissue) and atherosclerotic and unaffected arterial wall (n=40/tissue) isolated from CAD patients during coronary artery bypass surgery. The first step, across all mRNA signals (n=15,042/12,621 RefSeqs/genes) in each tissue, resulted in a total of 60 tissue clusters (n=3958 genes). In the second step (performed within tissue clusters), one atherosclerotic lesion (n=49/48) and one visceral fat (n=59) cluster segregated the patients into two groups that differed in the extent of coronary stenosis (P=0.008 and P=0.00015). The associations of these clusters with coronary atherosclerosis were validated by analyzing carotid atherosclerosis expression profiles. Remarkably, in one cluster (n=55/54) relating to carotid stenosis (P=0.04), 27 genes in the two clusters relating to coronary stenosis were confirmed (n=16/17, P<10-27and-30). Genes in the transendothelial migration of leukocytes (TEML) pathway were overrepresented in all three clusters, referred to as the atherosclerosis module (A-module). In a second validation step, using three independent cohorts, the A-module was found to be genetically enriched with CAD risk by 1.8-fold (P<0.004). The transcription co-factor LIM domain binding 2 (LDB2) was identified as a potential high-hierarchy regulator of the A-module, a notion supported by subnetwork analysis, cellular and lesion expression of LDB2, and the expression of 1, On the day of the defence day the status of this article was: In Press.

Synapses on dendritic spines of pyramidal neurons show a remarkable ability to induce phosphorylation of transcription factors at the nuclear level with a short latency, incompatible with a diffusion process from the dendritic spines to the nucleus. To account for these findings, we formulated a novel extension of the classical cable theory by considering the fact that the endoplasmic reticulum (ER) is an effective charge separator, forming an intrinsic compartment that extends from the spine to the nuclear membrane. We use realistic parameters to show that an electrotonic signal may be transmitted along the ER from the dendritic spines to the nucleus. We found that this type of signal transduction can additionally account for the remarkable ability of the cell nucleus to differentiate between depolarizing synaptic signals that originate from the dendritic spines and back-propagating action potentials. This study considers a novel computational role for dendritic spines, and sheds new light on how spines and ER may jointly create an additional level of processing within the single neuron. © 2008 Shemer et al.

Härtill 4 uppsatser

BACKGROUND By offering a comprehensive view of the molecular underpinnings of pathology, high-dimensional data have the potential to revolutionize the diagnosis and management of complex disorders such as coronary artery disease (CAD). To identify molecular phenotypes of CAD, we performed multi organ gene expression profiling of subjects enrolled in the Stockholm Atherosclerosis Gene Expression (STAGE) study. METHODS Atherosclerotic and unaffected arterial wall, liver, skeletal muscle, and mediastinal fat biopsies were obtained during coronary artery bypass grafting from 114 well-characterized CAD patients. RNA samples were isolated, and 278 transcription profiles were obtained using Affymetrix HG-U133_Plus_2 GeneChips. RESULTS The most prominent molecular phenotype of the CAD patients was represented by 733 genes in mediastinal fat, which were involved in extracellular matrix organization, response to stress and regulation of programmed cell death. Other aspects of this phenotype were shared with liver (e.g., oxidoreductase activity), skeletal muscle (insulin-like growth factor binding), and atherosclerotic arterial wall (cell motility and adhesion, fatty acid metabolism). In addition, the activity of 400 genes exclusively in mediastinal fat was associated with the extent of coronary stenosis and atherosclerosis. Immune-cell activation in mediastinal fat defined CAD patients with poor blood glucose control and prolonged hospitalization. CONCLUSIONS The molecular phenotype of mediastinal fat appears to be central in CAD and should be useful for early identification of CAD risk.