27 October 2010
Reviewed by Felix Grant
The 13th edition of GenStat (formally version 13.2, after a service release) continues recent development trends, with broad based development of reach and function centred around an emphasis on biosciences, but plenty of appeal for broad scientific application.
This approach finds particular expression in a new menu, one of several progressive changes under the top level Statistics drop down, collecting tools for qualitative trait loci analysis under the heading QTL (Linkage/Association). This is more than an interface metaphor, though: the GenStat server stores and maps information about QTL data structures in a very intuitive and productive dedicated centralised space for common access by the tools grouped here.
At import of traits, maps or marker data, structure names can be added to this space and the contained structures viewed with a utility which contains addition, import, export and shortcut buttons. Provided they have been specified, QTL tools automatically enter structure names into the appropriate fields when their menus are opened, and hide from display. Right-clicking on the available data list calls up a shortcut to change between data space and full program data display.
Using the new data space are a number of new or improved QTL tools. Eigenanalysis and linkage disequilibrium options make a welcome appearance alongside comprehensive access to resampling via the bootstrap procedure. Subsetting by genotype and marker joins summary statistics options for markers and between traits, as does imputation of missing marker scores. Analyses of multi-trait linkage and (single or multiple environment) single trait analyses branch off association or simple and composite interval, respectively, branch off from the QTL Analysis submenu. Instant import and export of data in Flapjack format is added to GenStat’s famously broad file handling gamut.
There is, as always, a good spread of development across various areas of practice. There are other menu developments besides QTL (regression, mixed models, multivariate analysis, time series), further extending the options for quick and intuitive exploratory control. The spreadsheet interface sees several productivity tweaks, and new ways to manipulate matrices are a welcome addition to data handling. Graphics refinement continues with new enhancements to axis labelling, colour handling, symbol plotting and text box control.
At a more radical level new directives, functions and procedures extend fundamental capability, some of which are covered by or related to comments above (QTL gaining QASSOCIATION, QEIGENANALYSIS, QLDDECAY, QMKDIAGNOSTICS, QMVREPLACE, QMVAF, for example) while others are in different directions.
A group of server directives fit, display and predict from multi-layer perceptron neural networks. These work impressively with minimum interference, but there is all the scope for fine tuning you would expect from a GenStat provision, the fit itself offering ten options and eight parameters. Particular favourites of my own include extensions of classification. BCFOREST constructs a random forest (with, once again, a rich set of control options) while BCIDENTIFY and BCDISPLAY then do exactly what you might expect of them.
BIDENTIFY identifies specimens using a key or tree built using BKEY, BCLASSIFICATION, or BREGRESSION. Calculation of regression models and percentile or quantile calculation across multiple variates, and spatial point spectra, are other highlights amongst many. And I haven’t even mentioned the range of incremental development within existing facilities.