Replicated data strategy

W. Smith and T. R. Forester, Parallel macromolecular simulations and the replicated data strategy . Comp. Phys. Comm., Vol 79, no 1, 52-62, 1994. 

W. Smith and T. R. Forester, Daresbury Laboratory Preprint DL/SCI/P873T Warrington, U.K., 1993. Parallel Macromolecular Simulations and the Replicated Data Strategy. I. The Computation of Atomic Forces. 

Smith W 1992 A replicated data molecular dynamics strategy for the parallel Ewald sum Comput. Phys. Commun. 67 392-406... 

W. Smith, Comput. Phys. Commun., 67, 392 (1992). A Replicated Data Molecular Dynamics Strategy for the Parallel Ewald Sum. 

One of the important features of MD is that its performance can be significantly improved through code parallelization. Commonly employed strategies for parallelizing an MD code include (1) the atom decomposition (or replicated-data) scheme, (2) the force decomposition scheme, and (3) the... 

The level of validation to be undertaken must be chosen considering scientific and economic constraints. All data have some value, and results from the development phase can all be pressed into service for validation. Separate planned experiments might lead to better and more statistically defensible results, but when this cannot be done, then whatever data are at hand must be used. The best use can be made of experiments to be done by understanding what is required. For example, in a precision study, if the goal is to know the day-to-day variability of an analysis, then duplicate measurements over 5 days would give more useful information than 5 replicates on day 1, and another 5 on day 5. The strategy would be reversed if variations within a day were expected to be greater than between days. 

There are several strategies and tools that can be used to assess, and subsequently minimize, such sampling-based errors in the calibration data. The use of replicate samples in the calibration experiment can help to detect noise issues around instrument... 

The main point of this contribution, however, is the relation between data set sizK and reliability of biomarker identification. As expected, all the methods become less efficient as the number of biological replicates decreases, but even in these conditions the use of PLS-DA and the f-test offer effective biomarker identification strategies. This observation is fxmdamentally important in all studies where it is impossible to acquire more samples, and suggests that small sample sizes can still allow reliable selection of biomarkers. 

Furthermore, recent statistical advances have expanded the repertoire of tools with which to analyze data tfom these designs. For example, hierarchical linear models (J. E. Schwartz, Warren, Pickering, 1994), random regression models (Jacob et al., in press), or pooled cross-sectional time series (Dielman, 1983) allow for the partitioning of inter-individual and intra-individual variability from a number of different sources. Complemented by iet a a/vric techniques that allow for the examination of multiple dependent variables (Cohen, 1982), these methods offer many data analytic strategies for multivariate, replicated, repeated-measures, singlesubject designs. Several of these techniques are illustrated in the next section. 

---