CANCER COMMONS: BIOMEDICINE IN THE INTERNET AGE - Collaborative Computational Technologies for Biomedical Research - page 169

knowledge, and resources. By coordinating their efforts, a network of personal-
ized virtual biotechs can systematically explore the opportunity space of
targets and leads and avoid unnecessary replication of experiments. By virtue
of such sharing, a network of personalized virtual biotechs can achieve unprec-
edented economies of scale and acceleration by leveraging knowledge about
targets and leads across diseases.

11.3.2 Cancer Commons as a Macroscale N - of - 1 Adaptive Trial

All patients whose treatment is being managed within Cancer Commons are,
in effect, participating in a huge adaptive clinical trial testing the molecular
disease model and leading to changes in the model resulting from sources such
as personalized virtual biotech. One can think of this as a macroscale (across
the whole Cancer Commons community), N - of - 1 (each patient is being treated
through personalized genomic medicine), adaptive trial. Let us see how and
why this is needed and how it works.
Treating cancer (as is the case for most diseases) is, in computational par-
lance, a very high dimensionality search problem. The number of potential
hypotheses about the causes and associated treatments of the disease is huge,
especially when complete genomic profi les and combinational therapies are
considered—exponentially larger than the number of patients that could rep-
resent each possible combination of factors.
Classical large-scale clinical trials are inappropriate for systematically
exploring very high dimensional spaces where, as we have seen, there are
potentially a huge number of differences at the genomic level and a huge space
of possible treatments. One reason for this is that a large-scale controlled trial
is very likely to have a heterogeneous population of patients with functionally
different genomic tumor profi les. As a result, it is possible that a treatment
that failed statistically in a large-scale controlled trial actually helped a small
subset of the patients who have a specifi c genomic profi le, but these cases were
mixed together with many others who were not helped by the treatment. To
make matters (much!) worse, the most effective treatments will probably
involve combinational therapies and dosages, resulting in a nearly infi nite
number of potential treatments, even if we only consider approved drugs.
Computer scientists call the situation “the curse of dimensionality” [5]: There
are nowhere near enough patients to explore a space of this size using classical
clinical trials.
Even though the raw dimensionality of genomic profi les crossed with treat-
ments is so high that there cannot be enough patients to sort out genomic-
level treatment effectiveness precisely, a great deal is known about cancer
treatment from large-scale controlled trials, from research using in vitro , in
vivo, and in silico models, and from the experiential knowledge of the treat-
ment community. The molecular disease model approach taken by Cancer
Commons posits that a great deal is actually already known about how to
treat many genomic subtypes of cancer, but this knowledge is not gathered