Talk Series „Computational Medicine“


Die Fakultät für Informatik, die Fakultät für Mathematik und die Medizinische Universität Wien laden alle Interessierten herzlich zur Talk Series „Computational Medicine“ ein.

11.10.2019, HS 16 (Oskar-Morgenstern-Platz 1, 1090 Wien)
Bernhard Knapp (International University of Catalonia, Barcelona)

"Mapping algorithms for next generation sequencing data"
I am not an expert in next generation sequencing data. Never the less I would like to seize the opportunity to step out of my comfort zone and acquaint myself with this topic explaining the application of two very well-known computer science algorithms for the mapping of next generation sequencing data: I will explain the application of "hashing" and the "Burrows– Wheeler transformation" in simple terms and discuss the beauty of these two algorithms in the context of mapping of next generation sequencing data. By this way I hope to show my ability to teach interdisciplinary subjects as well as my ability to learn quickly.

"Computational prediction of T-cell receptor sequences for patient specific (cancer) antigens"
The affinity of T-cell receptors (TCRs) to peptide/MHC (pMHC) complexes (e.g. a specific cancer antigen) can be improved by introducing mutations in TCR sequences. Testing a large number of mutant TCRs experimentally is expensive and time consuming. A computational high-throughput approach could test such large numbers of TCRs in short time and at little cost. This method would be of high value in many areas of immunology such as cancer, viral infections and vaccines. However, to date no such method exists. Current computational methods are either computationally too expensive or not accurate enough. Here we propose to extend our previously developed MOSAICS protocol (Knapp et al. 2017 Bioinformatics; Sim et al. 2012 PNAS) for the prediction of TCR binding. MOSAICS implements a unique combination of hierarchical Monte Carlo, coarse-graining, stochastic chain closure, and (local) temperature annealing cycles. By this way MOSAICS will allow the prediction of association, dissociation, and binding of TCRs with peptide/MHC using structural modelling while non-essential degrees of freedom are restricted. MOSAICS is orders of magnitudes faster than conventional molecular simulation approaches and achieves high agreement with experimental data as we have shown before in multiple studies. This new MOSAICS extension together with deep learning approaches will for the first time allow computational optimisation of TCRs and antibodies on a patient specific basis. This will be of high value for precision medicine as for example in chronic viral diseases, cancer neo-antigens, orphan TCRs, and fundamental insights into autoimmune pathogenesis.


18.10.2019, HS 17 (Oskar-Morgenstern-Platz 1, 1090 Wien)
Doron Levy (University of Maryland)

"Modeling the development of drug resistance"

In this lecture addressed to students in mathematics, computer science, and medicine, we will overview basic concepts in modeling the development of drug resistance. We will start with a brief description of the Luria & Delbrück (LD) analysis for the development of antibiotic resistant bacteria. This work received the Nobel prize in medicine (1969). We will then demonstrate how the LD analysis was extended by Goldie and Coldman to model resistance to chemotherapy in cancer.

"Mathematical modeling the role of the immune response in leukemia"
Modern targeted therapies have significantly improved the treatment of chronic myelogenous leukemia (CML). Yet, most patients are not cured for undetermined reasons. In this talk we will show how mathematical methods can be applied to model the immune response to CML. Along the way, we will discuss cancer vaccines, drug resistance, and cancer stem cells. We will demonstrate how mathematical methods can be used to integrated clinical and experimental data in order to guide treatment options.


22.10.2019, SR 15 (Oskar-Morgenstern-Platz 1, 1090 Wien)
Spiros Denaxas (University College London)

"Methods for translational research using electronic health records"
Electronic Health Records (EHR) are a rich source of information on human health and disease and offer substantially larger phenotypic depth and breadth compared to traditional research-driven studies and clinical trials. This talk will demonstrate the opportunities and challenges of using EHR for translational research and cover methods for creating and validating disease phenotypes using complex heterogeneous data.

"Data-driven approaches for extracting actionable knowledge from electronic health records"
Electronic Health Records (EHR) are data generated and captured during routine clinical interactions and offer a wealth of information on disease and it's progression. EHR however are noisy, complex and biased and require a significant amount of pre- processing in order to be transformed to research-ready datasets. The talk will cover data-driven supervised and unsupervised phenotyping methods and approaches for deriving clinically meaningful and actionable knowledge from EHR data and showcase the opportunities and challenges of such data for research and care.

23.10.2019, HS 2 (Oskar-Morgenstern-Platz 1, 1090 Wien)
Ruth Pfeiffer (National Cancer Institute, Bethesda, Maryland)

"The survival function: The Kaplan-Meier nonparametric estimate"

In many medical studies, the main outcome of interest is the time from disease diagnosis to an event such as death. Such event times are often called “survival times”, but the ideas also apply to other types of events. If all patients could be followed indefinitely, all the survival times would be observed, and the empirical distribution function F(t) of survival times could be computed. Its complement, S(t)=1-F(t) is called the estimated survival distribution. However, some patients may drop out of the study (are lost to follow-up) or the study ends before their survival times are observed. Their survival times are said to be "censored"; one only knows that the censored survival time of interest exceeded the censoring time. I will describe a non- parametric estimate of S(t), called the Kaplan-Meier (KM) curve, which is widely used in medical and other scientific applications. I will show how to test the difference between two KM curves at a given time point.

"Building, updating and validating risk models for clinical practice and public health"

Statistical models that predict disease incidence, disease recurrence, or mortality following disease onset have broad public health and clinical applications. I introduce various definitions of "risk". Of great importance are models that predict absolute risk (also called 'cumulative incidence' or 'crude risk'), the probability that an individual who is free of a given disease at an initial age, a, will develop that disease in the subsequent interval (a, t]. Absolute risk is reduced by mortality from competing risks. I discuss approaches to building risk models by combining data from various sources, and how one can update models when data on new risk factors become available. Before a risk prediction model can be recommended for clinical or public health applications, one needs to assess how good the predictions are. I will present some general criteria for model assessment and some criteria I developed to assess the usefulness of such a model to make screening decisions. Accommodating missing model predictors is an area of ongoing work that I will present. I also highlight some challenges to improving model performance. To illustrate many ideas, I use a risk model I developed to predict incidence of breast cancer in US women.