DCBIOSTEC 2021 Abstracts

Full Papers
Paper Nr: 2

Experimental Evaluation of a Gas Sensors Array for the Identification of Complex VOCs Mixtures in the Breath of Patients by Pattern Recognition Techniques


Justin Martin

Abstract: Lung cancer is one of the deadliest form of cancer in Europe, being the first and second cause of cancer death respectively for men and women. This high death toll has to be blamed on the lack of obvious symptoms in the early stages of the illness. Current diagnostic methods tend make the screening costly and difficult to organise at a large scale. Asymptomatic subjects and people in remote areas are rarely tested overall, leading to late discovery of the cancer and poor survival chances. There is therefore a need for a diagnostic method that could be used remotely while being simple enough to be used with little prior formation. Gas sensor arrays have properties fitting for the task. This thesis aims at creating and testing a sensor array in order to build a benchmark on which one can compare the discriminative power of different arrays. Several tasks will be performed simultaneously: The first is the establishment of a standardized test method of the metrological characteristics of commercial thick film sensors as well as experimental ones, and their qualities within a sensor network. The second is the integration of experimental sensors into a prototype gas sensor array consistent with the final purpose of the device. The third is the validation of the test method with the prototype electronic nose, which requires the reproducible synthesis of reference gas mixtures. It is also planned to use real breath from healthy persons and cancer patients as validation of the benchmark’s conclusions. The last task is about the processing and analysis of data and the identification and classification of samples in order to obtain a measurement of the array’s discriminatory power. This thesis is part of the PATHACOV research project, funded by Interreg France-Wallonie-Vlaanderen.

Paper Nr: 4

Motility Analysis and Classification of Lipid Droplets in the Cytosol of Living Cells


Jonas Schurr, Julian Weghuber, Peter Lanzerstorfer and Stephan Winkler

Abstract: In this paper, we present an analysis tool for motility classification and analysis of lipid droplets. We provide a fast and easy-to-use semiautomatic workflow for a comprehensive motion analysis. It includes unsupervised image processing algorithms for the detection, tracking, feature extraction and further analyses. We developed a simple unsupervised but robust detection algorithm. It is based on a convolution filter which can be adapted for high reusability. No preprocessing is required. A neighborhood-based approach for the tracking allows the extraction of important metrics describing the trajectory of each droplet. With these features we provide a customizable threshold-based classification to simplify further examinations. Additionally, the mean squared displacement is calculated at each time point to allow a determination of the motility mode. Based on these values an estimation of unknown parameters can be performed to provide additional information about the diffusion. With our approach we speed up, simplify and further automatize the information extraction process of lipid droplet analyses.

Paper Nr: 5

Machine Learning Models for Predicting Disease-related Biological Interactions


Ivan Carrera

Abstract: Drug repositioning is a process that, by means of identifying new targets for a specific drug, changes its initial purpose. Analyzing biological interactions, i.e., relationships between drugs and their targets, or associations between cell lines and genes, is an effective approach for drug repositioning, and for predicting side effects of drugs in patients. Drug repositioning based on the analysis of scientific literature is growing at a quick pace, becoming a high-impact big data problem. Health big data can benefit from being analyzed using machine learning strategies for extracting and summarizing information, and automated processing will allow scaling up the analysis with a much broader set of data. The main goal of this work is to develop a set of computational models to automatically identify, classify, and discover new biological interactions, namely new drug-disease interactions, as well as drug-cell line interactions. Machine learning and data mining techniques are used to identify interactions over a set of pharmacological and biological databases, and scientific literature. Data for these analyses will be obtained from scientific literature and open databases, such as ChEMBL, DrugBank, KEGG, PubMed, OMIM, etc. Once interactions are classified, we will introduce a prediction model to identify new possible effects of known drugs. Preliminary work has shown promising results for characterizing cell lines and diseases, and for discovering new interactions between chemical compounds and cell lines, as well as between chemical compounds and diseases.