The last decade has seen an explosion in the fields of genomics, from the Human Genome Project announcing their successful sequencing of the first complete human genome, to the rapid advances in sequencing technologies that have been exponentially lowering the price and time of data acquisition, to the development of companies such as 23andMe that, for just $99, can tell you things about yourself that may be able to predict diseases in advance. In fact, for $199, you can get part of your ancestry traced by National Geographic. I, a Chinese Canadian, may just find out that I’m 6% Ashkenazi Jewish or 4% Sub-Saharan African. So with such great advancements in genomics, why isn’t this more of a thing? What are the challenges that genomics needs to overcome in order to make the transition into widespread commercial success?
1. Biology is the world’s most ridiculously complicated science.
When 23andMe was issued a warning, along with four other companies, for not being able to submit evidence that their results reliably indicated the medical outcomes they predicted, it didn’t surprise me, but not at all because I thought 23andMe was a bad company – quite the opposite. It was because if they could really submit that evidence, it would mean that they had achieved the holy grail of genomics: the perfect mapping of changes in DNA to changes in the way our body functions. Me and all of my colleagues would be promptly without a job. For those who haven’t touched this since high school, DNA is a template for RNA, which is a template for proteins, which builds all sorts of things in cells. But perhaps you didn’t know that DNA has a 3D structure and is wrapped in proteins called histones, which can release the DNA when interacting with specific proteins, which in turn can take their signals from RNA that don’t turn into proteins. I could go on forever, but the idea is that DNA has a ton of information, but after each layer of processing, more and more of that information is scattered, like pool balls, into other components. Retracing those components is all I do for a living, and trust me, it’s no easy going.
2. The price of failure could mean someone’s life.
One thing that differentiates genomic prediction from predicting who is most likely to buy a summer dress off of ModCloth is that genomics speaks to human health, which deals with life or death situations, literally. Angelina Jolie got a much publicized double mammoplasty after learning that she had a mutation in BRCA1 associated with breast and ovarian cancers. Mutations in BRCA1 followed by family history of cancer is certain enough, even in my books, to warrant drastic action, but most mutations are not so clear cut. Jolie has a likelihood of cancer quoted to her in the high eighties, but sometimes it’s in lower amounts. What do you do if your genetic screen tells you that you have a 25% likelihood of cancer? What happens if you have a 46% chance of something you never end up getting and you end up suffering a disease that the screen missed as a false negative? Each individual’s biology is so distinct that we currently have no way to predict how people will react to most drug treatments. From DNA to protein, there is huge scattering of information. From proteins to human bodily function, just imagine that the pool table got a thousand times bigger and now there are pool balls that you haven’t heard of and cannot see.
3. The feedback loops for biological information are longer and more indirect.
How long does it take to actually find out if your algorithm is correct? And how do you know if your algorithm performed well? These are two of the most pressing questions for any statistician, and biostatisticians once again have the short end of the stick. Health matters are things that take years, if not decades, to find out, in a lot of cases. IVF treatments can go for half a dozen rounds before a baby is conceived, and if you’re predicted to have a disease, it may be 20 years before that disease actually manifests itself. For the statistician collecting data, it can be challenging to find out years and years later what worked and what didn’t, especially since they are dealing with life-altering events. Most studies can only be funded for a certain amount of years, which means that anything occurring out of that window of time is lost to the ages. Ultimately, even if someone is able to collect all of that data, there is no way to trace it back to the genetic mutation that supposedly predisposed the person to the illness. We can only average over a gigantic population of cooperating individuals and hope that the aggregate can shed a light into the individual.
So what does this mean for the future genomics predictive analysis? I know I’ve probably made it sound quite bleak, but that’s just what science people do (we’re not very fun at bars). This article was about the potential dangers of predictive analysis of genomics, but amazing strides that have been made since the discovery of the mechanisms of DNA information transmission, from the identification of mutations for a wide catalogue of genetic diseases (including the gene that makes you smell like a fish, which is the subject of an actual book) to the discovery of gene therapy techniques that may one day reverse such problems. As long as scientists exist, every problem under the sun is being slowly chipped away at, so one day we may very well be able to connect a person’s genomic profile in a deterministic way to their health outcome. And that day may be soon.
Pamela Wu is a graduate student and research assistant at NYU Langone Medical Center and an InSITE Fellow. Her research is on bioinformatics and genomics .