Experience of OpenStack Test Deployment and Comparison of Virtual and Real Cluster Environments

Introduction. A brief overview of the properties and architecture of one of the components of the National Cloud of Open Science prototype – the cloud platform OpenStack is given. The list of software and hardware components of the OpenStack test cloud environment and the sequence of actions required for the deployment of both OpenStack itself and the Slurm virtual cluster environment for portable, scalable, reproducible scientific biomedical computing are presented. The purpose of the paper is a description of the experience of test deployment of OpenStack to create a scalable computing environment for reproducible scientific computing using modern technological solutions, which can be applied to both cloud (OpenStack, AWS, Google) and cluster platforms (Slurm). Results. The structure of the created test containerized (using Singularity technology) biomedical application, which contains modern software and libraries and can be used in conventional and cloud virtual cluster environments is briefly described. The results of a comparative test of this application in the virtual cluster environment Slurm under the control of OpenStack and in the node of cluster SKIT-4.5 in the V.M. Glushkov Institute of Cybernetics of the NAS of Ukraine are given. Information on solving the problem of finding the optimal in terms of saving resources scaling parameters for the developed application in two comparable cluster environments is given. Some features of the use of these cluster environments are clarified, in particular, a comparison of the dependence of the application speed on the number of parallel processes for two cluster environments is presented. Empirical data are presented in graphical form, which illustrate the nature of the load on the OpenStack server and the use of RAM on the number of parallel processes. Possibilities of portability between the specified cluster environments, scaling of calculations and maintenance of reproducibility of calculations for the offered test application are demonstrated. The advantages of using OpenStack technology for scientific biomedical calculations are pointed out. Conclusions. The described example of test deployment and use of OpenStack gives an idea of the requirements for the necessary technical base to ensure the reproducibility of scientific biomedical calculations in cloud and cluster environments. Keywords: cloud technologies, reproducible calculations, cluster platform.

An Intelligent Control Model of Credit Line Computing in Intelligence Health-Care Systems

Technologies such as machine learning and artificial intelligence have brought about a tremendous change to biomedical computing and intelligence health care. As a principal component of the intelligence healthcare system, the hospital information system (HIS) has provided great convenience to hospitals and patients, but incidents of leaking private information of patients through HIS occasionally occur at times. Therefore, it is necessary to properly control excessive access behavior. To reduce the risk of patient privacy leakage when medical data are accessed, this article proposes a dynamic permission intelligent access control model that introduces credit line calculation. According to the target given by the doctor in HIS and the actual access record, the International Classification of Diseases (ICD)-10 code is used to describe the degree of correlation, and the rationality of the access is formally described by a mathematical formula. The concept of intelligence healthcare credit lines is redefined with relevance and time Windows. The access control policy matches the corresponding credit limit and credit interval according to the authorization rules to achieve the purpose of intelligent control. Finally, with the actual data provided by a Grade-III Level-A hospital in Kunming, the program code is written through machine learning and biomedical computing-related technologies to complete the experimental test. The experiment proves that the intelligent access control model based on credit computing proposed in this study can play a role in protecting the privacy of patients to a certain extent.

Obtaining leaner deep neural networks for decoding brain functional connectome in a single shot

Neuroscientific knowledge points to the presence of redundancy in the correlations of brain’s functional activity. These redundancies can be removed to mitigate the problem of overfitting when deep neural network (DNN) models are used to classify neuroimaging datasets. We propose an algorithm that removes insignificant nodes of DNNs in a layerwise manner and then adds a subset of correlated features in a single shot. When performing experiments with functional MRI datasets for classifying patients from healthy controls, we were able to obtain simpler and more generalizable DNNs. The obtained DNNs maintained a similar performance as the full network with only around 2% of the initial trainable parameters. Further, we used the trained network to identify salient brain regions and connections from functional connectome for multiple brain disorders. The identified biomarkers were found to closely correspond to previously known disease biomarkers. The proposed methods have cross-modal applications in obtaining leaner DNNs that seem to fit the data better. The corresponding code is available at https://github.com/SCSE-Biomedical-Computing-Group/LEAN_CLIP.