The screening problem here is to search through many pairs of variables
that have been measured on a common sample, identifying pairs that are
likely to have a structural relationship on some subset of the sample.
An example is the screening of thousands of medicinal compounds, whose
biological activity (Growth Inhibition) has been measured on 60 cancer
cell-lines curated by the National Cancer Institute (NCI-60), against
expression levels of hundreds of microRNA (miR), which have recently
been measured on the same NCI-60 sample. It is believed that miRs play
a role in the chemo-sensitivity and chemo-resistance of cells, but it
would be extraordinarily expensive to conduct dilution assays for all
compounds on cells with each mir individually silenced or stimulated.
Thus there is a need for screening a manageable number of compound-miR
pairs for further experimentation.
The statistical approach involves a modification of Kendall's tau
test for independence, where observations are re-ordered according
to their components of concordance. The re-ordered components form
a path whose distribution is simulated under the null hypothesis of
independence to establish an acceptance region. Excursions outside
this region indicate highly associated subsets. This tau-path method
has better power than Kendall's test when the alternative involves
a small but highly associated subpopulation. In this type situation,
it also performs better than several other ad hoc methods, including
tests based only on the longest monotone subsequence, and tests involving
transformations to a location shift problem.
This is joint work with Li Yu, Paul Blower and Shili Lin.
Meet the speaker in Room 212 Cockins Hall at 4:30
p.m. Refreshments will be served.
