Skip to content

Upcoming

______________________________________________________________________________________

Dimarts 4 de juliol (15:00 - 17:00)

Sala de reunions FME (Campus Sud)


Carles Serrat + Núria Pérez + Esteban Vegas: Survival Kernel Analysis: State of the art and next steps.

Abstract:

Sparse kernel methods like support vector machines (SVM) have been applied with great success to classification and (standard) regression settings. Existing support vector classification and regression techniques however are not suitable for partly censored survival data, which are typically analyzed using Cox's proportional hazards model. The available methods for performing survival analysis in high dimension data sets are assessed and an example of application to microbiota data and survival time is described.


Héctor Sanz Ródenas: Support Vector Machines for Survival Analysis: Methods and Variable Relevance (Assaig Tesi Doctoral).

Abstract:

The primary objective of this thesis has been to propose new SVM for survival data approaches by extending the binary outcome approach to time-to-event, and proposing new methods to visualize and rank the relevance of the predictors involved in these survival-SVM methods. Therefore, applying these methods to the Mal067 data to help understanding the immune responses induced by the RTS,S vaccine and by natural exposure to the parasite that are associated with malaria risk or protection.


Assajos xerrades ISCB.

Yovaninna Alarcón: Data Science in HIV studies.

Abstract:

The goal of this talk is two-fold. First, we present an overview of my thesis proposal and the main results found so far. Second, we would like to discuss some difficulties and challenges faced during the beginning of the project.

The thesis project aims to cover the following two large blocks: one block is rather applied and consists of developing new techniques to work with data related to HIV (like data from trials with therapeutic vaccines and enzyme-linked immunospot (ELISPOT) assays), and the other block is more theoretical, consisting of joint models of survival and multi-omics data.

The development of immunologic interventions to control viral rebound in HIV infection is a major goal of the HIV-1 cure field. Therapeutic vaccination with “kick and kill” strategies has been proposed to control viral replication after discontinuation of Antiretroviral Therapy. Kick and kill strategies mean to wake up or activate latent reservoir cells (kick) and to teach the cells of the immune system to recognize these activated cells and kill them.

Besides, concerning HIV-related data, there is little information so far, on data obtained from ELISPOT assays. The ELISPOT assay is an immunoassay that measures the frequency of cytokine-secreting cells at the single-cell level. A single cell forms a colored “footprint” (spot) on the bottom of the well representing its secretory activity.

The main idea of the analyses of the data from therapeutic vaccination and ELISPOT assays is to identify which kind of variables are correlated with clinical parameters and if some of them can be important for a prediction model.

On the more theoretical arm, to identify biomarkers it is essential to study different omics layers and associate them with the survival of patients. Different approaches have been developed here1,2, but nothing in the context of HIV. Because of this, one aim of my thesis is to study existing and develop new joint models for survival and omics data.

Collecting these ideas and establishing a doctoral thesis project in the field of biostatistics has its difficulties, from which also the great challenges arise. As an example, one difficulty is to manage a common vocabulary between the statistician and the clinician, and then to be able to find mutual profit. The clinician wants to answer a question and the statistician is interested in developing a new methodology to give the response, this is how a great network of multidisciplinary cooperation is born.


Marta Bofill Roig: Sample size derivation for binary composite endpoints.

Abstract:

Composite binary endpoints (CBE), defined as the union of several binary endpoints, are frequently used as the primary endpoint in a clinical trial. The specification of the treatment effect on the composite endpoint requires the information of their components and the degree of association between them. We summarize the treatment effect on the composite endpoint by means of the odds ratio and show that is determined by six parameters including the degree of association between components, the event proportion and the corresponding odds ratio of the individual endpoints.

The purpose of this talk is to explore sample size formulations for CBE in terms of the marginal odds ratios, the event proportions and the degrees of association. We discuss the influence of each of the parameters of the composite in the required sample size. While anticipated values for the marginal parameters are often easier to guess and most of the sample size formulations depend on those, anticipating the degree of association between the components is a much harder task. Within the framework of multiple co-primary binary endpoints, several authors have addressed the influence of association on sample size [Sozu2010]. However, approximations to the needed sample size of a CBE with partial or null knowledge of the correlation are limited. We aim to develop alternative formulas for the computation of the sample size for CBE.

 

 ______________________________________________________________________________________