What's New in Version 4.3?

• Linear, Rpart and Nnet Models - many improvements to the ModelAble interface for regression and classification tools are in this release. Some options have been removed and others added. Independent variables in all modeling routines are now always scaled to a common min-max(0-1) numerical scale prior to fitting. A second R-package, neuralnet, is now available as an option in the interface for neural nets. This library allows for 1 or 2 hidden layers and features a generalized weight analysis that can assist with interpretation of the model. Bayesian models which factorize the independent variables are now available with rpart and neural net logistic models.
• Improvements to Graphics - the interfaces used for plotting run, assayrun and dataframe objects have been enhanced with added functionality. BubblePlots reflecting 4-5 "dimensions" of a multi-variate dataset are available as a new plot type. In many cases, scatterplots and their pair-wise implementations now have a density2d option that allows for a more detailed appreciation of variable distributions where numbers of observations are high. Where possible, custom options for user annotation of graphical output has been added as a convenience. The default dpi for graphics has been changed from 96 to 192 pixels per inch. Nine additional color schemes have been added for use with plots that employ color gradients.
• Help System - tooltip help for ScreenAble's various client interfaces has been greatly expanded. In all cases where related tools share a common core interface, different settings will update these tips dynamically. The online help system is refreshed at application start and a PDF version can be downloaded from the main help menu.

Introduced in Version 4.2

• Linear, Rpart and Nnet Models - the R-language is widely used for building statistically valid linear and logistic models with experimental data. Quantitative models can be employed both to better understand sources of variance in the screening processes and to search (in silico) for interesting diversity to screen. Model building also plays a central role in QSAR research efforts. Data Mining, Machine Learning and Artificial Intelligence are all terms used to promote model building with large datasets. At ScreenAble Solutions, we agree with those who prefer to call these activities applied statistics. We are excited to make these tools available to our customers and will continue to refine them in future releases.
• Import support for RDS data files - the TextFile Viewer tool has been re-named to the DataFile Viewer to reflect an increase in its utility. It still supports comma and tab delimited text files, but now can read the compressed binary files (*.rds) used extensively in the R-Environment. This format has long been supported by the data export routines of ScreenAble. The ability to load them directly from *.rds makes this format useful for efficient storage and retrieval of working datasets.
• File support for Bayesian models - introduced in 4.1, the Bayesian Manager now has file support in addition to database storage. A text-based file format (*.sbm) has been developed to support this feature. The data processing "rules" for database models are unchanged. The emphasis in the database paradigm is a single screen or test. File based models provide support for using aggregated data or even external results to build models. In either case, any molecular diversity loaded using the SDFileViewer can be used with these models for predictive analyses.

As always, ScreenAble Solutions is committed to providing its Customers with the best technology available for their Discovery Screening analytics. Watch this space for future developments!