The Problem of Perception

Your interpretation of results depends on more than just the results.

Steven Wiley
Mar 1, 2009

There is a common perception among young students that the surest path to resolving scientific controversies is to design a clever experiment, one that will definitively resolve conflicting hypotheses. However, I have found that most scientific controversies do not revolve around specific experimental data, but instead are disputes over data interpretation.

Data interpretations depend on a scientist's underlying assumptions and worldview. For example, a molecular biologist might think of protein expression as an outcome of mRNA levels, whereas a biochemist might think in terms of synthetic and degradation rates. Both are right, of course, but each might expect different reasons for a change in the amount of a protein. Our perspective and assumptions regarding how living systems work defines us as biologists, which is why arguments over interpretations can get so nasty. If another scientist disputes the validity of your viewpoint, it can impact your reputation as well as your...

I experienced a dispute over data interpretation in the late 1980's that turned somewhat contentious. My laboratory was investigating the role of tyrosine kinase activity in the endocytosis of the EGF receptor in collaboration with Gordon Gill at the University of California, San Diego. By using a series of kinetic measurements, we concluded that kinase activity was necessary for ligand-induced endocytosis and down-regulation, whereas a competing group thought that it was not. What was interesting about the dispute was that the primary data gathered by both groups were essentially identical. The main difference was in how the data were interpreted.

We were calculating receptor endocytosis rates on a per receptor basis whereas the other group used a per cell basis. In the case of receptors with kinase activity, EGF binding accelerated endocytosis 10-fold, which caused cell-surface receptors to drop 10-fold. Thus, the net rate of endocytosis per-cell (rate x receptors) did not change in response to kinase activity, making it appear to have no effect on receptor internalization.

It was hard to convince the other group that differences in a simple rate calculation would make a big difference in what they concluded from the data. However, we were working from a computational model of endocytosis that allowed us to try out different sets of assumptions and see how they would affect the system's behavior. The other group felt that our computer model was a poor substitute for their own scientific intuition regarding what was happening. As they stated at the time, "(Our) results are in agreement with the results of Chen et al. (Cell, 59:33–43,1989), but in conflict with their interpretation. By applying a simple mathematical model (J Cell Biol, 107:801–10, 1988), the lack of down-regulation of kinase-negative mutants was misinterpreted as an indication for reduced endocytosis."1

Over time, our view prevailed and it is now generally accepted that the kinase activity of the EGF receptor is required for ligand-induced endocytosis.2 Interestingly, our view was vindicated not because people came to accept our use of computational modeling, but because our hypothesis was more successful in predicting subsequent experimental results. Scientists don't generally care about who is right or who is wrong in a dispute. They want a conclusion that can help predict their own experimental outcomes. Science is built brick by brick from ideas and concepts that can lead to the next successful series of experiments and concepts. If an idea doesn't support the next brick, it is discarded. It's natural selection in science.

Scientific disputes seem inevitable in any career, but mine gave me a keen appreciation of the need for caution in accepting simple interpretations of the behavior of complex systems. In science, we do not gather facts. We make observations. Our interpretation of observations is only as good as our assumptions and conceptual frameworks.

The ability of a simple computer model to correctly interpret a seemingly non-intuitive result was also quite revealing. It convinced me that as biology becomes even more complex, we will need computational models even more to help us out. They can not only let us see past our inherent interpretational biases, but can also be used to design experiments to test our concepts—a foundation for building a truly predictive biology.

Steven Wiley is a Pacific Northwest National Laboratory Fellow and director of PNNL's Biomolecular Systems Initiative.


1. A. Ullrich, and J. Schlessinger, "Signal transduction by receptors with tyrosine kinase activity," Cell, 61:203–12, 1990. 2. A. Sorkin, and L.K. Goh, "Endocytosis and intracellular trafficking of ErbBs," Experimental Cell Res, 314:3093–106, 2008.

Interested in reading more?

Become a Member of

Receive full access to digital editions of The Scientist, as well as TS Digest, feature stories, more than 35 years of archives, and much more!
Already a member?