Harvard Medical School is running a crowd-sourced Network of Enigmatic Exceptional Responders to see if Big Data can help win the War on Cancer (kicked off by President Nixon in 1971). There is an interesting WBUR story about the project.
Given the amount of damage that computing has done to health care via incompatible and impossible-to-use electronic medical records perhaps computer nerddom can redeem itself with a cure for cancer?
Dude, dude, dude. Gas fumes, just say no to them. The war on ‘cancer’ is a war on a ill defined biological reality that is a natural and unavoidable counterpoint to being multicellular. If people die ‘cancer free’ it is because they die before they could not make that claim (and if someone comes with some retarded story of some oldie dying 100+ of other than cancer, a plague of statisticians on your house!). Also, a ‘cure’ for ‘cancer’ is a biochemical agent, not some data in a computer. Human genomics has produced 0 so far (the biggest results in healthcare genetics are BRCA 1 and 2, both predating the genome bollox). Epidemiology? smoking is bad, not exactly new news. The data drive is bullshit, but it fills many pockets so why not? it is easy to sell the rubes the idea that big data will do whatever.
Please note, in case we have epidemiologists, statisticians, computer folks, and other pseudoscientists in the audience: some ‘association’ is data is worth diddly squat unless one can make a robust (say, 90% chances of being right? at least?) prediction about the future. Science is about predicting what will happen, not magically explaining what might have happened. Yes, a lot of predictions are empirically falsified, but that is the point. Explaining some data with magic thinking is a cargo cult. ‘Cancer’ needs biochemical agents, and the willingness to say, cases, controls, here we go, someone will bite the dust. Experiments fellas, not big data.
They need really detailed data about the genetics, treatments, & lifestyles of the patients over time, to detect enough differences between the successes & failures. People aren’t willing to share they spend all day sitting around commenting on blog posts instead of exercising, even to doctors.
Maybe I am missing something. 100 different individuals (responders). To potentially 100 different treatments, for potentially 100 different types of cancers. Cancers which, by the definition of “response” are either no longer present or substantially changed, rendering tumor testing all but meaningless. In any case, based on the selection criteria they are clearly not looking for commonalities between the tumors or the treatments.
They are requesting a stool sample on which to perform “genetic testing.” This seems to be a play on recent observations that the gut microbiome can influence response to cancer therapies (for example, http://www.sciencemag.org/news/2017/11/your-gut-bacteria-could-determine-how-you-respond-cutting-edge-cancer-drugs). Of course, their microbiomes may have changed significantly since their response due to a change in diet (not uncommon for cancer patients), a course of antibiotics (also not uncommon) or other factors. But, if they “find” something, a booster for a specific gut microbe to improve patient response could be a big money maker, however effective it winds up being. And possibly could be sold under “right-to-try” prior to clinical trial testing, or even as a supplement – a probiotic, for which no testing is required. Hopefully this is not their plan, and they will pursue the rigorous route. But this avenue is open. Buyer beware.
You’d be surprised what we can do with N-of-1 studies. See the successes in the Undiagnosed Disease Network for example where each case is often unique: https://undiagnosed.hms.harvard.edu
Thread drift warning!!!!! Phil, we have not seen a post about your medical school insider in quite a while.
“Big data is a lot like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.”
It is all the rage now especially in pharma where I work. Having more data and integrating it all in an easy to access data structure is worth the trouble. But the potential is so overhyped right now, that everybody is bullshitting if they think it will cure cancer and other diseases. The data we have and plan to have through these initiatives is so far from being comprehensive enough to do anything “Big”. There is so much uncovered terrain, it’s the equivalent of giving 1000 people each a flashlight in the dark to cover 100000 acres, each one submitting a paper on each of their ‘findings’ in the fields then doing that for a couple of decades, until one day you give them all walkie talkies, and then you call it Big Data.
Second what @GermanL wrote.
IBM Watson with their “big” data and AI effort is a fine example of a failed project (I know this because I worked on the project). There is more marking and hype than real usable result. And don’t get me started on Amazon’s Alexis or Apple’s Siri or Google’s Assistant or Microsoft’s Contact et. al. All those are old technology running on newer faster hardware.
All the stuff you hear about today with data-mining, classification, taxonomy, and even AI, et. al. is nothing new. I worked on all of this stuff going back to 1985 when data was still on CD’s.
Exactly what George said, this stuff is old news, just regurgitated again, now that we can store petabytes instead of kilobytes. Databases, Machine Learning, AI, neural networks, etc, are decades old. Now it’s Big Data, Data Lakes (oh brother), Deep Learning, etc etc
Part of the problem with the lack of success is that there are too many idiots who want to use the technology without understanding it (doctors, biologists not train in bioinformatics), and whose expectations are wildly out of touch with what is actually possible. See:
https://www.technologyreview.com/s/607965/a-reality-check-for-ibms-ai-ambitions/
Third what @GermanL and @George A. wrote
Anybody that has done graduate level control systems and signal processing knows that all of this AI, machine learning at etc. is just dynamic optimization and Fuzzy logic. Today, we just have fast computers and lots of memory, to try out these algorithms, some of them from the 1960’s, on big data sets.
We still do not have the fundamental understanding of how intelligence works for true AI.
A good source today for information on fuzzy logic and machine learning Dr. Bart Kosko
http://sipi.usc.edu/~kosko/
His book, Fuzzy Engineering, released in 1997 is really good.
https://www.amazon.com/Fuzzy-Engineering-Bart-Kosko/dp/0131249916/ref=sr_1_1?ie=UTF8&qid=1530302673&sr=8-1&keywords=fuzzy+engineering
Pavel, I haven’t heard of Kosko in years. “Neural Networks and Fuzzy Systems” from 91 was great. A few years after that came out a co-worker attended one of his lectures. At the Q&A somebody from the audience accused him of being entirely unscientific: his paper was readable, correct, and had enough information in it that they were able to recreate the results- clearly Kosko wasn’t an academic!