Maths becomes biology's magic number

This so much:

Cookbook approach to statistics without insight leads to terrible results, much data effectively thrown away - very costly data $millions worth.

I have some complex medical conditions and it is just terrible to see highly trained specialists who are nothing more than cookbook executors. Once you are off the beaten path they are at a complete loss.

The rampant statistical illiteracy in the medical world from general practitioners to researchers make this era of big data a huge risk.

is a molecular biologist and jokes that her medical practitioner friends know almost nothing about how the human body works.

> Memorisation

> Cookbooks

> Linear regression, maybe

I agree largely with @GarrisonPrime's thoughts regarding the systematic nature of biological systems. I also share the same sense of trying to understand the logic and intuition of the physiology I learned in the first two years of medical school while many of my peers were just trying to guzzle and regurgitate. I have to admit that I also fell into that mode as well at times, just due to the volume of material that was expected. But that's for another time.

I'm currently taking a year to work on research that is in systems biology/bioinformatics. While there are many things that I like about it, and I'm grateful that it presents an opportunity for me to continue learning about computer science, machine learning (been really getting into learning about Bayesian analysis this year) and biology I have to admit that this article sounds like it was written from the same vantage point that I stood on a couple years ago as I just started getting into this area of research.

The technology we have today to probe cellular systems is amazing and was literally the stuff of science fiction some 20 years ago, but it's not without its faults. This line from the article especially rang true to how I feel these days:

"But there's a problem. The vast data sets that give bioinformatics its power are also its Achilles heel."

The problem is that the systems biologists and bioinformaticists are most interested in are dynamic with complex regulatory systems that we don't have ways of measuring and most methods of measurement either completely destroy the system or alter its dynamics. In addition, it's akin to taking a snapshot of how the system is behaving at one instance in time or condition. Yet many times we are asked to use that information in a way that's akin to trying to describe the dynamics of an entire motion picture from 2 or 3 photos. And those photos are greyscale. Take for example mRNA-sequencing, a type of data that I work with frequently. It's trying to measure the amount of gene product that a cell or cells have at one point in time (basically trying to get a measure of how much geneX the cell is trying to produce). While it is an interesting measure and can give some insight into how the the cell may be adapting to different conditions, those measurements alone tell us almost nothing about the regulation behind those differences, which is the thing we really want to understand. It's a bit like seeing oil on top of water and then trying to infer the complex dynamics of geophysics that are occurring on the ocean floor. Not saying that it's not useful at all, and can help direct your attention to the next interesting thing, but I think that many people overestimate how informative the data is. And then there are still a lot of technical issues but that is a discussion for another day.

The other main point I want to make is that for all the data you think we have now about these biological systems, it's like a snowflake on top of the iceberg. Even many of these large consortium projects like ENCODE have relatively small amounts of information if you want to learn about some transcription factor or cell type that isn't one of the top 10 most well known or studied. And how many of those datasets out there are really lacking in good quality control, and then there is the politics of sharing data in an academic/research environment that is so competitive getting a job (that you will have to continue to work like a madman/woman at) is like winning the lottery.

OK, I don't want this to descend into a full blown rant. Main points - it's still really exciting, and it's a great time to have intersecting interests in medicine, math and computer science. Just that the tech we have to work with right now is still a bit nascent and expensive. I think there will be a point where systems bio and machine learning will revolutionize how we understand biology. We're just not quite there yet.

On a side note - where I do see a lot of potential right now where computer science and machine learning can start to make an impact is more on the clinical side of medicine and using ML to learn from the vast stores of EMR data. But that's also another discussion for another day. If you read this far, here is a smiley face, and have a nice weekend :)

> An astonishing amount of biology is rather mechanistic, systematic, and logical.

I disagree vehemently. Or, to be more precise, it is very likely that this is true in reality, but in terms of the dominant way it's viewed in both clinical and research practice, it's very much not.

The average paper's findings are qualitative, not quantitative, in nature. Of course, numbers, statistics, etc, will be used -- but they will be used to get p-values to support a qualitative conclusion, rather than to work towards a quantitative, systematic understanding of biology.

There are exceptions, of course -- a decent amount of biochemistry is quantitative enough -- but when you get to topics like cell signalling, "cell biology", etc, it is almost never quantitative.

I have a favorite question I like to use in seminars when someone is presenting findings along the lines of "protein A regulates protein B". The question is: "Among the various regulators of B, how important is A? Is A a key regulator of B, or simply one of many?". Or, even more simply, "What is the most important regulator of B?" I invariably get blank looks, as if this is an unreasonable or impossible question to answer. Of course it is, as long as your models are qualitative. The point of the question is to get the speaker to think about that, but so far, no luck.