There are too many stories to relate from Beyond the Genome 2012, held at Harvard’s New Research Building last week. Many of them are beautifully synthesized in Isaac Kohane’s notes from the meeting (start here and press “Prev” to view talk notes in order).
Instead of re-hashing well told tales, I will comment on what I found to be one of the most poignant statements of the meeting, which came from Jim Lupski of Baylor College of Medicine. “Scientists have no idea how limited clinical work is,” he said, “clinical phenotyping,” he clarified. “Diagnostic codes are often designed for payment.” Most of the meeting was upbeat, chronicling the successful use of next generation genomic data to identify disease-causing mutations. But speakers also touched upon the challenges that must be overcome in order to raise the identification rate from ~25% (which Susan Plon, also of Baylor, reckons to be the success rate of their exome sequencing program). Analytical challenges related to identifying copy number variation (CNV), long insertions and deletions (INDELs), rearrangements and the like were noted. As were the inherent difficulties associated with studying phenotypes that are heavily influenced by environmental factors or dictated by the products of multiple genes.
But Dr. Lupski highlighted a less-discussed impediment to these studies: the problem of defining the phenotype. In what way is an individual considered diseased or simply beyond the boundaries of normal phenotypic variation? Too wide a definition increases the chance of a “single” phenotype being caused by more than one genetic loci and has the potential to bury disease-causing variants beneath background thresholds. Too narrow a definition may exclude affected individuals from being included in a cohort, reducing the power of that group to uncover relevant mutations. Testing at multiple layers of phenotypic classification (from narrow to wide) may help to resolve this matter and reveal previously uncharacterized causative mutations. But first, clinical phenotyping must achieve the depth and degree of standardization to permit such experimental manipulations later on. I leave the thorny question of how such standards might be implemented to those with experience navigating the difficult landscape of clinical research. But until research-defined metrics replace insurance code-based metrics in clinical phenotyping, the full potential of personalized medicine cannot be realized.