Over lunch we were discussing random numbers. If you look at random patterns and you compare it with a big enough dataset, you will find statistically significant correlations. In fact, some people claim that random numbers can predict the future, and even predicted 9/11. More importantly you can use random or non-random biological systems to predict the stock market and get rich, provided you have rats and know how to use excel... The latest paper in nature biotech used high-resolution, non-invasive imaging techniques (ie: CT scan) , to quantify 138 "visual" traits. By comparing this dataset with tissue microarray of nearly 7000 genes, the authors were able to train their software to be able to "reconstruct" the transcriptome of a tumour simply by looking at it. Now to a certain degree, it is intuitive: if there are lots of blood vessels, there probably is a high expression of angiogenic molecules, and cell shape is somewhat linked to the cell type. However, being able to deduce over 80% of the transcriptome by imaging only 28 traits seems dubious. Still, this is a good news for "personalized" medicine: better imaging, of things like tumour blood vessels, means better predictions for chemotherapy and more accurate prognosis...
Tuesday, May 22, 2007
Random numbers
Over lunch we were discussing random numbers. If you look at random patterns and you compare it with a big enough dataset, you will find statistically significant correlations. In fact, some people claim that random numbers can predict the future, and even predicted 9/11. More importantly you can use random or non-random biological systems to predict the stock market and get rich, provided you have rats and know how to use excel... The latest paper in nature biotech used high-resolution, non-invasive imaging techniques (ie: CT scan) , to quantify 138 "visual" traits. By comparing this dataset with tissue microarray of nearly 7000 genes, the authors were able to train their software to be able to "reconstruct" the transcriptome of a tumour simply by looking at it. Now to a certain degree, it is intuitive: if there are lots of blood vessels, there probably is a high expression of angiogenic molecules, and cell shape is somewhat linked to the cell type. However, being able to deduce over 80% of the transcriptome by imaging only 28 traits seems dubious. Still, this is a good news for "personalized" medicine: better imaging, of things like tumour blood vessels, means better predictions for chemotherapy and more accurate prognosis...
Subscribe to:
Post Comments (Atom)
Podcast
2 comments:
Here's an abstract from the guy at Princeton that has the random number generators around the world that predicted 9/11.
Here is a great CBC piece on the shutdown of his and other non-mainstream science research at Princeton (aka PEAR) (audio: unfortunately in real audio) but a piece that makes you wonder about what was in the drugs in the 70's.
Also the now defunct PEAR website is good for those wondering about this flavour of hippie research.
Interestingly, I have a book on my shelf (one of the vast great unread) by Francis Crick (also a big fan of the LSD) called "The Astonishing Hypothesis" (1994), described as "a fascinating argument that consciousness and what has long been called the soul are now accessible to scientific investigation"...Flipping through it kind of makes me wish I would have time in the next century to read it...oh here's a nice Einstein quote at the start of Chapter 15 ("Some Experiments"):
"Through purely logical thinking we can attain no knowledge whatsoever of the empirical world"...
Hmmmm... is Albert Einstein himself giving us a coded message to drop some LSD?
Post a Comment