Systems Problem, Systemic Solution

I had the privilege of attending the Mathematics of ICBP conference in Tampa last week as well as a meeting regarding an upcoming non-profit start-up with the goal of finding new treatments using a very high-throughput testing apparatus.  The most important lesson learned from these meetings is that cancer is a systems biology problem spanning many scales, with applicability to multiple disciplines, and whose ultimate solution is going to involve the marriage of theory and experiment.

Indeed, from my work I am learning that the models we use--random forest, elastic net, linear regression--are all actually quite good and in fact ultimately similar in their predictive power.  Instead, we suffer from having relatively few data points relative to the number of parameters we might be interested is.  That is, while we may have dozens of clinical features and thousands of copy number variations and gene expression levels, we have knowledge of very few actual patients.  It gets worse, though, as there are concerns as to whether some of the data we have is even on the "real" cancer.  Below, I list the major issues and open problems in understanding cancer.

• We do not understand how to link microscopic behavior of cancer to the macroscopic behavior.  That is, while we may gather a great deal of copy number and expression data points, we don't fully understand how that connects to the patient-scale phenotypes:  metastasis, recurrence, and prognosis.
• Because of their wild growth rate and minimal restrictions on DNA mutations, tumors exhibit surprising heterogeneity.  This heterogeneity allows the cancer to evolve resistance to whatever drugs are applied to it through natural selection.  How can we study these heterogeneous sub-populations of tumor cells?
• Some cancers may originate from cancer stem cells which compose a small fraction of the total tumor.  The effectiveness of treatments may be gauged based on their effect on benign daughter cells while the true desired target would be the very mobile stem cells.  How can we learn to kill a cell that we may be unable to even isolate for study?
• When you take a tumor out of a person and put it in culture or into a mouse for study, the nature of the tumor somehow changes.  The cells within the culture become visibly different from their progenitors in human.  How can we say that lessons from the mouse and in vitro models of cancer will be applicable to treatments in the human environment?

It's not all problems, though!  I learned about some new and exciting developments in cancer systems biology as well:

• TCGA is going to release protein expression profiles within the next year!
• The Gene Expression Commons will soon give an 'absolute' reference point for gene expression data!
• Dr. Carla Grandori has shown the effectiveness of a high-throughput screening platform in finding new therapies for difficult-to-target cancers.

The way forward as I see it is simple:  this is a systems biology problem, and to tackle it we need more data, open data, and more integration of databases and models from theorists, experimentalists, and clinicians.  Systems biology is complex, but ultimately understandable if we can just put the pieces together.