Mining molecular contexts of cancer via in-silico conditioning.

Cell maintains its specific status by tightly regulating a set of genes through various regulatory mechanisms. If there are aberrations that force cell to adjust its regulatory machinery away from the normal state to reliably provide proliferative signals and abrogate normal safeguards, it must achieve a new regulatory state different from the normal. Due to this tightly coordinated regulation, the expression of genes should show consistent patterns within a cellular context, for example, a subtype of tumor, but the behaviour of those genes outside the context would rather become less consistent. Based on this hypothesis, we propose a method to identify genes whose expression pattern is significantly more consistent within a specific biological context, and also provide an algorithm to identify novel cellular contexts. The method was applied to previously published data sets to find possible novel biological contexts in conjunction with available clinical or drug sensitivity data. The software is currently written in Java and is available upon request from the corresponding author.