Reproducible research
Being able to reproduce scientific results is extremely important. We make great efforts having the analysis methods in the aroma suite produce the exact same results over generations of updates. For instance, for each release we run a large set of redundancy tests to assert the correctness of results and that methods are backward compatible.
Extensive testing
Before each new release, we run a large number of system and redundancy tests on real data to validate that the new code is backward compatible and does not break existing functionality. The tests cover a large number of chip types and statistical methods. The complete set of tests is available in the following directory:
path <- system.file("testScripts", package="aroma.affymetrix")
Reproducibility of existing implementations
In addition to this, most of the standard methods reproduce the results of the original methods/implementations, e.g. RMA. Here is a list of documents showing how well we can reproduce existing methods:
- RMA (background, normalization & summarization)
- gcRMA (background, normalization & summarization)
- RmaPlm() and affyPLM's log-additive summarization
- Reproducing SNPRMA in the oligo package
- CRLMM genotyping (100K and 500K)