I have to commend the father for such a valiant effort.
However, shouldn't he have been able to automate part of the process, if he was manually comparing printouts to the Ensembl reference genome?
The job was daunting: The printouts contained data for approximately 20,000 base pairs, and there was no feasible way to automate the hunt for variants. After putting in a full day of consulting and then getting the kids to bed, Rienhoff would retire to his attic and spend hours checking Beatrice's adenines (A), thymines (T), guanines (G), and cytosines (C) against the Ensembl reference genome.
Who knows the real story. This could be a case of the author taking journalistic liberties, or simply misunderstanding the process.
My understanding is that a simple diff would take care of 90% of the work by eliminating all common base pairs. Then, from the results, his job would be to cross-reference each of the mutations against the gene database. Any unknown mutation could be the one he's looking for.
Perhaps this is actually what he did, and the "no feasible way to automate the hunt" refers to the cross-referencing step, not the differencing step.
Still though, reading that part almost made me email him saying, "I'll find you a way to automate it!".
The usual way to do a gene diff is by aligning two sequences with a program like ClustalW (free, and packaged with Debian now). That shows the differences and positions in a format other programs can read. But I don't know how much annotation the lab gave him, and annotating a pile of introns and exons does take a long time, even with the right tools -- the diffs aren't necessarily helpful until the rest of that information is in place.
I wish this guy a lot of luck. Whether he has a chance or not, I respect him immensely for not "standing on the sidelines".
The story really hits close to him for me because I have one of the disorders they narrowed out for the poor girl. Like it said, 90% of the battle for surviving some of these is just knowing what you have. It's an overwhelming feeling of frustration and helplessness to have your life 100% at the mercy of a dysfunctional healthcare system. At the risk of using a gratuitous nerd metaphor, this guy's writing his own OS instead of using Windows.
One question is if it is as simple as that. The human genome is immensely complex; much like raw object code (and compacted and self-modifying at that). The number of active genes are far too small for the amount of variation, so maybe she just got a bad run of genes. Unfortunately, that would make Reinhoff's search useless, because he would never find that one misplaced gene, and missing the four or ten that acted in concert to bring this about.
It probably isn't as simple as that, but in this case the downside of failure was nothing more than wasted time while the upside was a possible insight into a disease that had eluded diagnosis.
What is more troubling to me is the part of the article stating that he used this to justify canceling the MRI that had been advised by a doctor. I don't have much of a problem with people looking to unconventional medicine when areas that have been well researched have failed. This is especially true when the unconventional methods might have some basis for working and just haven't been studied enough. The problems often start, however, when people turn away from established medicine in favor of the untested methods. I know that he had reservations about that MRI beforehand, but using the gene sequencing to justify canceling the MRI seems to be a step in the wrong direction.
I find it rather depressing. Comparing the harm something like this could do (cheek swab and blood sample) and the potential benefit against some of the work scientists are performing on their own, healthy, children (there was a post on here about this a few days ago but it seems my searching abilities haven't returned from their lunch break yet) I really don't understand why he hasn't received more support.
I wonder how useful the algorithms used in those online 20 questions games such as 20q.net or akinator.com would be for diagnosing these sorts of rare diseases. The 20 questions games seem to be able to guess even extremely obscure objects or characters, so it seems as though it might be a useful tool for doctors in cases where the diagnosis is difficult. Obviously, a tool such as this would not be able to actually make a diagnosis itself, but it seems like it might be an extremely useful aid to doctors who are facing rare and obscure conditions.
The wikipedia article seemed to indicate that these sorts of systems don't see much use outside of limited fields in medicine. Do you know why that is? It seems to me that the problem should be very similar to that faced by those 20 questions games, yet they seem to do fairly well on a wide variety of topics.
Also, to differentiate diagnoses will frequently take tests and things that normal people don't have access to. A lot of things cause headaches, a lot of things cause nausea (damn near everything does, it seems like), etc. From an information theoretic point of view, it's hard to get enough bits of information from a normal person to diagnose anything. From a Bayesian point of view, it's hard to get enough evidence to diagnose a rare disease starting with low priors, but a system that only diagnoses common diseases is of limited value.
In a limited domain it can do well, or even outperform a doctor (who is also limited by having a large domain, that problem is actually fundamental to the problem space not just the intelligence applied), in general it can't.
I've half-self-diagnosed myself with Celiac disease (wikipedia article is the best on it I've found), which is in the running for one of the most under-diagnosed diseases currently known, precisely because the symptoms are relatively vague and quite variable ("Symptoms include chronic diarrhoea, failure to thrive (in children), and fatigue, but these may be absent and symptoms in all other organ systems have been described." - can it get any more vague? yes, but only barely.). This is an extreme, but there's enough medical conditions like this to make the general case very difficult to deal with.
Self-diagnosis is a whole other ball game, and it's a dangerous game to play. While a good expert system could help doctors significantly, in the hands of the general public it could also cause great harm.
Self-diagnosis delays consultation with an actual expert. In addition, many symptoms (especially neurological ones) can only be seen or elicited by a trained third-party.
PS: I have a friend with Coeliac disease, so I know something about it. It is hard-to-diagnose conclusively, but that's often not necessary. A key feature of the disease is that treatment, elimination of dietary wheat gluten, is harmless and usually causes drastic improvement. In my friends' case, it didn't, but that's an exception. I strongly recommend talking to a doctor and dietician about it.
I mean no disrespect, but that "half-self-diagnosed" isn't "self-diagnosed" for a reason. I don't need your approval of my decisions. I was simply using that as an example of something hard to diagnose. As you say, it's even hard to diagnose with medical resources, so it's not as if there's a choice. The whole "don't self-diagnose" speech is misplaced when diagnosis is not reliably available!
Beatrice's underlying problem, if genetic, will probably either be a new mutation, or will be caused by having two recessive alleles inherited from each of her parents. If the former, then comparing her genome with her parents' genomes would (I would guess; IANA geneticist) isolate it pretty quickly.
However, shouldn't he have been able to automate part of the process, if he was manually comparing printouts to the Ensembl reference genome?
The job was daunting: The printouts contained data for approximately 20,000 base pairs, and there was no feasible way to automate the hunt for variants. After putting in a full day of consulting and then getting the kids to bed, Rienhoff would retire to his attic and spend hours checking Beatrice's adenines (A), thymines (T), guanines (G), and cytosines (C) against the Ensembl reference genome.