Biomedical Engineering Reference
In-Depth Information
One area in which commercial vendors still have a key role is the of use
of data analysis tools in regulated environments. Extensive software
validation is a costly exercise and currently commercial software vendors
take this responsibility and worry away from the customer, naturally in
exchange for licence fees. For those interested in this area, Chapter 21 by
Stokes discusses this in much more detail.
In more research-based areas, open source solutions may have an edge
over commercial software due to the openness of algorithms. There have
been recent moves to make data analysis more transparent by embedding
algorithms inside publications using tools such as R and Sweave [58].
There has been much discussion about 'reproducible research' [59, 60],
where not only the data but also the methods used for analysis are freely
available. This could be a valuable contribution to the aim of being able
to reproduce complex 'omics analyses.
Open source is certainly a challenge to the commercial software world
and we have seen the rise of many new business models, such as open
community development with support, consultancy and training being
run as commercial services. One area of relevance to Metabolomics is the
provision of online services [61]. How open and free these type of services
are likely to remain if massive transfers of data occur, remains to be seen.
Commercial cloud computing solutions may emerge; but data security is
of great concern to commercial organisations. I believe most commercial
researchers would be very uneasy about transferring sensitive metabolomics
data, particularly in highly commercial areas such as disease diagnosis or
regulatory studies.
Lastly in the data analysis area, it appears so far only commercial
vendors have been able to produce truly interactive data analysis tools.
Products such as SIMCA-P, Unscrambler and JMP allow interactive
plotting, data exclusion, fi ltering, transformation and many other
manipulations. These operations are essential to any large-scale data
analysis task where several rounds of quality control and data cleanup
must be performed. To date, most of the data analysis tools in the open
source are script-based and offer little opportunity for truly interactive
analysis. Two packages which are attempting to make R more user-
friendly and interactive are Rattle [62, 63] and GGobi [64, 65] but to date
these have focussed more on 'machine learning' for business applications
rather than chemometrics. (In general terms, machine learning algorithms
require many more observations and fewer variables than are encountered
in metabolomics, hence the preferred use of Chemometric methods, which
are specifi cally designed for 'wide' data sets.) It will be interesting to see if
any open source interactive tools will be developed in the future.
Search WWH ::
Custom Search