Národní úložiště šedé literatury Nalezeno 4 záznamů.  Hledání trvalo 0.00 vteřin. 
Segmentation of phase contrast images in multi epitope ligand cartography (MELC) for image quantification at the single cell level
Mívalt, Filip ; Taschner-Mandl,, Sabine (oponent) ; Mehnen, Lars (vedoucí práce)
The Multi-Epitope Ligand Cartography (MELC) technique enables microscopy-based visualisation of multiple cellular compartments by using immunofluorescence stains. A MELC data processing pipeline as previously established in-house within an ongoing research project, providing biologists with a tool for quantitative antibody signal analysis. The pipeline, therefore, allows phenotype characterisation of cells present in bone marrow aspirates from neuroblastoma patients. The antibody signal assignment to the plasma membrane of single cells is based on nuclear segmentation and region growing, but only approximates the real cellular shape. This approach is particularly error-prone when applied on touching or overlapping cells due to an ambiguous assignment of a single antibody signal to multiple cells. This error, subsequently, propagates to single-cell level features, thereby possibly influences ensuing phenotype classification or quantification. Hence, the segmentation of phase contrast images acquired simultaneously with each fluorescence image and visualising the whole cell (including cytoplasm and nucleus), is required to provide the pipeline with accurate segmentation masks representing the entire cell. We implemented an automated strategy employing a Mask R-CNN network to segment these phase contrast images. The algorithm achieved an overall object-level F1 score of 0.935 and a pixel-level F1 score of 0.868 when training with only a small annotated dataset. The trained model was implemented into the existing MELC data processing pipeline. Moreover, we provide an annotated dataset comprising 54 phase contrast images of bone marrow cytospin preparations containing an overall number of 1,940 cells. The implemented Mask RCNN model enables to study single cell-level features using segmentation masks representing cells predicted from phase contrast images and therefore improves an automated quantitative analysis of bone marrow samples for children’s cancer research.
Quality assurance of RNA-Seq workflows with spike-ins controls
Drozd, Tomáš ; Turk, Andreas (oponent) ; Mehnen, Lars (vedoucí práce)
Spike-in controls such as External RNA Controls Consortium (ERCC) or Lexogen‘s Spike-In RNA Variants (SIRVs) have become essential when it comes assessment of technical variability. Since the E0 SIRVs have identical concentration, variations in their estimated concentration can be used to infer the technical variability from single replicates. This is more economic than the standard approach, which estimates the technical variance from multiple replicates. The DESeq model, a standard tool for differential expression, was modified based on spike-ins information to estimate technical variability. Subsequently, the probability of a change in expression due to technical variability was obtained. A high variation between SIRV transcript read counts was discovered, giving rise to another approach based on estimation of variability for each trasncript separately. This innovative approach revealed better performance on datasets, where only technical variability was present for cross-condition analysis for a given number of replicates per condition. It was observed that increase in number of samples results in higher reliability for estimation. However, spike-ins, especially SIRVs, improved performance of analysis than estimation based on endogenous genes when a few replicates are available. Further reasearch is needed for normalizing technical varibility to estimate significant changes in biological variation.
Unsupervised Deep Learning Approach for Seizure Onset Zone localization in Epilepsy
Přidalová, Tereza ; Cimbálník, Jan (oponent) ; Mehnen, Lars (vedoucí práce)
Epilepsy affects about 50 million people worldwide, with one-third of patients being drugresistant and therefore candidates for an invasive brain resection surgery. Brain resection surgery candidates undergo invasive intracranial encephalography (iEEG) monitoring to determine the seizure onset zone (SOZ). Recorded data can span over weeks and need to be manually reviewed by a physician to assess SOZ. This process can be time-consuming and burdensome due to the vast amount of collected data. This work investigates utilisation of an deep autoencoder for unsupervised data exploration and specifically its ability to discriminate between SOZ and non-SOZ (NSOZ) iEEG channels. The data used in this thesis consists of iEEG collected from 33 patients in two institutes (Mayo Clinic, Rochester, Minnesota, USA and St. Anne´s University Hospital, Brno, Czech Republic - FNUSA) who underwent invasive presurgical monitoring. The autoencoder’s capability to discriminate between SOZ and NSOZ was evaluated using a self-learned embedded feature space representation of the autoencoder network. Autoencoder features were compared to previously established biomarkers for SOZ determination. Discrimination capability was evaluated for both autoencoder features and biomarkers using a Naive Bayes classifier and leave-one-out cross-validation. The achieved area under receiver operating characteristic curve (AUROC) was 0.68 for the FNUSA and 0.56 for the Mayo dataset. Performance in discriminating between SOZ and NSOZ electrodes was not significantly different between the investigated autoencoder features and previously established biomarkers. Selecting the better performing classifier for each patient increased the AUROC to 0.75 and 0.64 for the FNUSA and Mayo dataset, respectively. The results suggest that future approaches combining biomarkers and self-learning methods have a potential to improve the SOZ vs NSOZ discrimination capability of unsupervised iEEG exploration systems, and thus to enhance the surgical management of epilepsy.
Machine learning models for quantifying phenotypic signatures of cancer cells based on transcriptomic and epigenomic data
Koban, Martin ; PhD, Florian Halbritter, (oponent) ; Mehnen, Lars (vedoucí práce)
Since the advent of techniques capable of rapid acquisition of genomic data, it is one of the key challenges for researchers to interpret the results of such experiments in meaningful biological terms. In this work, we aim to exploit knowledge hidden in well-characterised transcriptomic and epigenomic data from publicly available sources to aid this interpretation. An integrated resource of chromatin accessibility data (from DNase-seq and ATAC-seq experiments) was created and pre-processed for downstream analyses, complemented by collections of public gene expression (RNA-seq) profiles. These datasets were used for training machine learning classifiers with two primary purposes. Firstly, for augmenting sample annotations by predicting undefined metadata labels in the training datasets. Secondly, for annotation of poorly characterised, unseen data to examine generalisation ability of the constructed models. We demonstrated that biologically relevant information was captured by the trained classifiers while technical artefacts were minimised. Thus, we validated that the proposed supervised machine learning approach can contribute to clarifying contents of cryptic transcriptomic and epigenomic datasets, particularly from the field of cancer research.

Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.