Forum:Hand-waving Multidisciplinary Aspects of Research Projects (Discussion)

Question

[Spencenerd]

Hi all,

Bioinformatics is a field that is particularly cross-disciplinary. That is, most BI projects involve elements of at least a couple of fields, most obviously Biology and Computer Science, but often others like Medicine, Statistics, etc.

While reading papers primarily based in one field, I have often found that the elements of other disciplines are "hand-wavy" or grossly oversimplified. Despite this, the papers usually seem to be founded well in science (at least from the perspective of the main discipline) and make positive contributions to the field, even when the methods utilized would be considered sub-par from the perspective of a member of a different field.

Some questions for discussion:

Is it acceptable for us to allow this sort of hand-waving in publications?

At what point is an oversimplified application of a cross-discipline harmful?

Can this type of research contribute to science?

Would it be overkill to expect these researchers to use state-of-the-art methods in all disciplines involved in a project?

What other questions should we be asking about this?

I look forward to the discussion about this topic.

[dariober]

Ironically I find this post to be overly hand-wavy and as such difficult to discuss. What do you mean by elements of other disciplines are "hand-wavy" or grossly oversimplified? Which other disciplines? And what makes them oversimplified? Can you give some examples?

[WouterDeCoster]

Reminds me of this tweet:

I think for my next talks I need to start glossing over the wet lab parts by saying "did some in vitro/vivo experiments" #vaam2016

— Kai Blin (@kaiblin) March 16, 2016

Answer 1

[Devon Ryan]

I think this post sort of misses the point. You don't need to use the latest and greatest methods, you need to use methods sufficient to support your conclusions. Papers should be relying on multiple lines of evidence (all with different biases) to support their conclusions. Even if some NGS/bioinformatics methodology is a bit sub-optimal, it's that combined with the other methodologies that matter. Insisting that everyone always use the latest and greatest methodologies at all times even when a conclusion is already sufficiently supported is wasting everyone's time and grant money.

Answer 2

[Alex Reynolds]

New class offered by UW, left here without comment:

Calling Bullshit in the Age of Big Data

Each of the lectures will explore one specific facet of bullshit.

Answer 3

[Petr Ponomarenko]

Science is all about discovering new things. The quality of tools and methods used for research has to be just good enough. In the field of engineering tools and methods used usually of much higher quality and precision, because mistakes there are much more costly. Nowadays many tasks in bioinformatics are moving from science to engineering. I feel the same as you mostly because of this.

Answer 4

[John]

Both computer science and biology are two fields that are pretty bad at managing complexity. Biology because it is so inherently complex that we get used to dumbing down concepts to the point of absurdity (like naming proteins/genes after that 1 thing they did that 1 time in that 1 specific context...) Computer Science on the other hand is half man-made and half objective truth, and it's difficult to distinguish what part of an algorithm is fundamentally complex, and what is just complex due to the implementation. A lot of work has been spent in Comp. Sci trying to manage complexity, but only in the past 5 years are people starting to really figure it out with practical solutions to these philosophical problems: https://www.infoq.com/presentations/Simple-Made-Easy

So as a result bioinformatics is just one of those fields where there's such a high % of unnecessary complexity mixed in with necessary complexity that it's a real endless rabbit hole if you want to try and explain something simple to someone. Just imagine trying to explain how reads are mapped in practice to a genome with an aligner like BWA. It would probably take over a week to explain to a biologist, no matter how eager they were to learn. And not because they're stupid, but because when BWA was written no one cared if you could explain how it worked to a biologist. All that matters is that it can map reads better than the next program.

That's why there's so much hand-waving. You want the details but the details are irrelevant 90% of the time because we just made them up. Why do you have to run "samtools index" before Picard will MarkDuplicates? because reasons.

EDIT: I've been a real downer the past couple of months. I'm sorry