Please activate JavaScript!
Please install Adobe Flash Player, click here for download


ISCB 2014 Vienna, Austria • Abstracts - Oral Presentations 17Monday, 25th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust allocated to toxic doses from sigmoidal, two from conservative and five from other approaches. Adding 10% to the prior probabilities had little effect. Conclusion: The CRM outperformed the 3+3 method. The underlying model of the dose-toxicity relationship influences the number of patients allocated to toxic doses. The sigmoidal approach was optimal by these cri- teria, although the two patients receiving the higher dose in the conserva- tive approach may give more confidence in the MTD.   C05 High-dimensional data analyses I C05.1 Assessment of positivity in ELISPOT assays based on FDR-type and mixture procedures I James1 1 Murdoch University, Perth, Australia   ELISPOT (enzyme-linked immunosorbent spot) assays are widely used in immunotherapy trials to assess whether T-cells of patients respond to im- mune stimulus. Reactive cell responses lead ultimately to spots on plates which can be counted and compared with the numbers of background spots resulting from corresponding unstimulated cells. Typically there will be a small number of replicates of both stimulated and unstimulated wells for each patient, with the assay repeated across a number of patients. Assessment of a“positive”response requires determination of a real excess of responses or spots in the stimulated wells compared with their back- ground. A number of ad hoc and statistical methods have been proposed to indicate positivity. Assessment of ELISPOT response is similar to the issue of false-discovery rate analysis in multiple testing where the aim is to determine those p- values associated with real effects. Here we consider adaptation of the FDR-type approach to ELISPOT analyses and compare it with an approach based on a novel mixture method in which the numbers of samples re- sponding positively are estimated directly without a binomial sampling assumption.The latter method appears to provide some robustness to the estimation over a range of parameters encountered in this setting. C05.2 Estimating individual peptide effects from pooled ELISPOT data P Ström1 , NJ Borthwick2 , T Dong2 , T Hanke2 , M Reilly1 1 Karolinska Institutet, Stockholm, Sweden, 2 Oxford University, Oxford, United Kingdom Background: In studies of the immune system, interferon-γ ELISPOT as- says are used to detect the T cell responses to peptides of interest. The peptides are placed in the wells of 96-well plates, either individually or in groups (“pools”) and after incubation with the blood cells to be tested, the responding cells appear as spots, with a high number of spots indicating that the peptide(s) stimulated a strong immune response. Objective: To develop and implement a method for estimating individual peptide responses from the ELISPOT responses from peptide pools and investigate its performance in a study where individual peptide responses were validated. Materials and methods: By regarding the individual responses of pep- tides in a pool as“unobserved”data, the problem can be solved using the EM algorithm. We model the response from any pool as a Poisson random variable whose expected value is the sum of the intensities of the individ- ual peptides that constitute the pool. We apply the method to data from 4 volunteers in a HIV-vaccine trial where 199 individual peptides were as- sayed in 80 overlapping pools and compare our estimates to the observed responses from the individual confirmatory assays. Results: The EM equations can be solved by simple matrix operations to yield the maximum likelihood estimates of the individual peptide re- sponses. Applying the method to the pooled responses from the 4 volun- teers, we demonstrated excellent agreement between the estimates and the individual validations.   C05.3 Estimation of antibody concentration from multiple dilutions data Y Xu1 , YB Cheung1 1 Duke-NUS Graduate Medical School, Singapore, Singapore In medicine and chemistry, measurement of concentrations usually in- volves calibration that maps the observed response to the underlying concentration level using the inversion of a standard curve. The Enzyme- linked ImmunoSorbent Assay (ELISA) is one such method that is com- monly used to measure antibody concentration. A problem in this type of technology is that an accurate measurement is obtainable only if the observations fall within the optimal, near-linear range of the standard curve. It is common to conduct a series of doubling or tripling dilutions of the samples, so that at least some of the diluted samples are within the optimal range. A single dilution may then be selected for statistical analysis. This common practice does not fully utilize the data from mul- tiple dilutions and reduces accuracy. We consider a weighted average estimation approach for fully utilizing the information from multiple di- lutions. Simulation results demonstrated the superiority of this weighted estimation approach over the conventional approach of analyzing a single selected dilution. We apply the methods to an experimental study of vac- cine candidates. C05.4 Mixed models for the analysis of brain magnetic resonance imaging data KJ Lee1,2 , DK Thompson1,3 , PJ Anderson1,2 , LW Doyle1,4 , JL Cheong1,4 , JB Carlin1,2 1 Murdoch Children’s Research Institute, Melbourne, Australia, 2 University of Melbourne, Melbourne, Australia, 3 Florey Neuroscience Institutes, University of Melbourne, Melbourne, Australia, 4 Royal Women’s Hospital, Melbourne, Australia Advanced Magnetic Resonance Imaging (MRI) of the brain often involves measuring specific quantities, for example volume or diffusion param- eters, in a number of different regions in both the left and right hemi- spheres of the brain. We explore approaches to the analysis of such data when we are interested in a comparison of brain metrics between two groups of individuals. We compare the standard approach, where each region in each hemisphere is analysed separately, to a mixed model ap- proach, where we combine all of the regional data into a single regression model including a random effect to allow for the multiple measurements on each individual.Within the mixed model approach we compare the use of a single or region-specific random effect and single or region-specific error term. The advantage of the mixed model approach is that it jointly estimates the differences in the quantity of interest across the regions.This is more efficient than conducting separate comparisons and enables an assessment of overall patterns of differences across the regions, thereby reducing the multiple testing and increasing the power to find a differ- ence between the groups compared with the standard approach. It also provides a direct assessment of whether the effect of group varies across the regions i.e. whether there is a region-specific effect of group. We ap- ply these approaches to data from the Victorian Infant Brain Study where we wish to compare the regional brain volumes at term-equivalent age between infants born very preterm and those born at term.

Pages Overview