IEEE Technology and Society Magazine - June 2019 - 55

synergetic understanding of cancer
evolution and the effect of combination drug therapies on the disease
is the cornerstone for developing
effective persona lized treatments, which can radically improve patients' well-being and
their quality of (work and social) life. By extension,
improving the treatment of patients indirectly enhances the quality of life for families, friends, and careers.
Moreover, personalizing effective therapeutic approaches reduces treatment duration, cutting down
healthcare monetary costs, which can be redirected
to other health and social services. Given that three
out of four U.S. families will at some point experience a family member suffering from cancer (http://
natamcancer.org/NAP_Native_ American_Priorities
.pdf), the potential impact of improved cancer treatment is of considerable socio-economic and organizational significance.
Paving the way for the development of new cancer
treatments requires identifying personalized drug combinations, which, by acting synergistically, will be able to
fight cell resistance in target therapies and ultimately
increase the patients' life expectancy and quality of life.
Studying the effects of drug synergies on resistant cells
is central to the problem of curing cancer. These phenomena are subject to the inherent complexity of biological systems and affected by the interplay between
different processes that occur at different scales. For
instance, they depend on the molecular mechanisms by
which individual cells can develop resistance to a particular drug [1]. These molecular mechanisms are complex
in their own right. But drug synergies' effectiveness also
depends on various types of dynamic processes related
to populations of cells. Examples include variability in
the gene expression profiles of different cells, which
gives rise to heterogeneous populations; competition for
resources such as space and nutrients; and interaction
or cross-talk between different cells [2]. Consequently,
multi-cellular systems' dynamics, such as tumor growth
and evolution, can only be understood by studying how
individual cells grow, divide, and die, and the interactions between the cells at the population level.
In-silico models are becoming powerful tools in the
fight against cancer, since they allow the combination,
in a controlled environment, of heterogeneous sources
of experimental data with prior biological knowledge,
towards a better understanding of the underlying mechanisms and the biological processes that determine
tumor growth, resistance to therapies, and effects of
drug synergies in cells. The multi-scale interplay of such
processes makes their modeling and simulation difficult. For instance, an average-sized tumor nodule
JUNE 2019

∕

contains approximately 108 -109 cells. Each such cell
needs to be modeled individually, thus its state must be
updated at each step of the simulation process. Moreover, modeling the dynamics of the entire system at the
cell-population level, and the evolution of various environmental parameters via partial differential equations
adds additional complexity to the task. The process of
simulating such a cellular system produces data at a
rate of approximately 100 GB/min. Therefore, in-silico
models of tumor evolution and resistant cells' emergence require immense computational resources, in
addition to extremely efficient data processing algorithms, operating on massive data streams.
This article presents the necessary architectural and
algorithmic apparatus for speeding up a) the repetitive
procedures employed by biologists in cooperation with
data scientists for modeling, setting up, and running
such simulations, and b) the extraction of simulation
outcomes so that drug combinations are interactively
applied. This way, the time for personalized and consequently, more effective, treatments is cut down.

Technological Challenges
Below, we summarize the technological challenges that
must be addressed for the development of new cancer
treatments.

Challenge 1
On one hand, utilizing large computer clusters, along
with high-performance computing (HPC) resources in
order to avoid stalling clock rates when dealing with
extreme-scale simulations, is of utmost importance. On
the other hand, more often than not, data analysts cooperating with life scientists have to resort to traditional,
batch processing methodologies, where a computer
cluster is treated as a black box. A user submits a job to
the cluster and patiently waits for the job to finish, with
little insight on what is being processed and where, what
data volumes are being moved around and at what cost,
how long the whole process will take, and whether it
could be optimized to run more efficiently. Often, what
lies at the end of this long and expensive process, is yet
another set of experiments. Consider, for instance, a life
science data analyst who needs to tune a vast number of
parameters for a complex machine learning model, or
identify potentially useful cell or population features,
with no alternative other than repeatedly running a different parameter/feature set configuration.

Requirement 1
Addressing this challenge requires developing novel interactive processing tools for extreme-scale data analytics. Such tools will enable life scientists in cooperation
with data analysts to iteratively pose queries and derive

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

http://natamcancer.org/NAP_Native_American_Priorities.pdf http://natamcancer.org/NAP_Native_American_Priorities.pdf http://natamcancer.org/NAP_Native_American_Priorities.pdf

IEEE Technology and Society Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - June 2019