6
$\begingroup$

This question was prompted by differential gene expression complex design no replicates.

While that question isn't very clear, it seems to be essentially asking for help on applying statistical methods to get a result. It isn't asking for help on a bioinformatics method, but for help on choosing the right statistical approach to analyze the results of a wet-lab experiment (RNAseq). If this were asking for bioinformatics methods for analyzing RNAseq data, that would be fine, of course, but if the OP is requesting for help on choosing the right statistical approach, do we consider that on topic?

Note that if the answer requires you to understand the underlying biology, then of course it should be on topic. I am thinking of situations where the details of the dataset are irrelevant and the question is one of pure stats which could be aplicable to any field.

$\endgroup$
10
$\begingroup$

I generally agree with @Ian Sudbery, but I would phrase this as I would prefer that the site's purview be "anything an expert bioinformatician might reasonably know that has at least a tangential relationship to bioinformatics". This would include statistics, biology, and some more general CS related questions provided they have some clear relationship to bioinformatics.

I should note that this makes it exactly the same as biostars, the scope of which some people don't like.

$\endgroup$
  • 1
    $\begingroup$ My feeling about the distinction about here and biostars is that biostars is explicitly aimed at beginners, where as I feel it would be useful if this could be a place where experts found it useful to ask questions as well as answering them. Questions where the answers wouldn't be completely obvious to anyone who had been in bioinformatics for a while, but required properly specialised knowledge about a sub-area, weather that was the stats of an unusual datatype, how to implement a particular algorithm efficiently, or a genome annotation feature particular to a specific species/assembly. $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:33
  • 3
    $\begingroup$ I think the scope of biostar is good. It is the other things that concern me more. $\endgroup$ – user172818♦ Jul 26 '17 at 13:38
  • $\begingroup$ I'm not sure how explicitly biostars is aimed at beginners, but it's certainly become that way. $\endgroup$ – Devon Ryan♦ Jul 26 '17 at 13:39
  • 3
    $\begingroup$ @IanSudbery ideally, we should be able to do both. I mean both help experts and help novices. I don't see the two as mutually exclusive. $\endgroup$ – terdon Jul 26 '17 at 13:40
  • $\begingroup$ DevonRyan: I'm sure I'd seen that somewhere, can't find it now though. @terdon absolutely, I'm just commenting on the idea that also supporting experts could be something that distinguishes us from biostars (even if biostars not supporting experts is by accident rather than design). $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:49
7
$\begingroup$

I think stats if very much on-topic. If we want the site to be useful for experts to ask questions and get informed answers from other experts, rather than a site where newbies are just instructed on how to install a piece of software, or where to find the limma manual, how to convert one file format or one gene id format to another (and other sites already fill this niche, then I think many of these will effectively be stats questions.

There is already an SE site dedicated to statistics, but in my experience questions that are about the stastics of bioinformatics rarely get helpful answers there. Partly I believe that this is because by the time you've explained the requisite bio background to understand the question, people have switched off. Here we get better engagement.

For example, I asked effectively the same question on here and on cross-validated.. After nearly 2 years the cross-validated answer has no useful responses, while the one here got answers very quickly.

Of course this is probably related to how you view bioinformatics, which I basically regard as a branch of applied statistics these days.

$\endgroup$
  • $\begingroup$ I am thinking about situations where the bio part of the question is irrelevant. if you need to understand the underlying biology to be able to answer the question, then it belongs here. No argument there. If, however, the biology is irrelevant and the question is one of pure stats, then I think it belongs elsewhere. I'll try and edit my question to make that distinction clearer. That said, I personally view bioinformatics as theoretical biology and absolutely not as statistics. Of course, that's because my background is in biology. $\endgroup$ – terdon Jul 26 '17 at 12:28
  • $\begingroup$ Take a look at my questions linked - its basically pure stats (how do a characterise the relationship between two spatially resolve count variable with high amounts of missing data). But unless you can explain why you are trying to do that, you don't get any answer other than "why on earth would anyone want to do that". What is theoretical biology other than mechanistic modelling (which I grant isn't stats) and statistical modelling. $\endgroup$ – Ian Sudbery Jul 26 '17 at 12:35
  • $\begingroup$ I'd say your question is on topic, yes. But that's because it is about a particular type of data whose characteristics you need to know in order to deal with it. In other words, a typical example of a case where you need to understand the underlying biology in order to provide an answer. Note that the answer you got is basically "use these standard bioinformatics tools" and not "use this stats method". (I won't answer the "what is bioinformatics" question here since it would derail the discussion; happy to chat about it though!) $\endgroup$ – terdon Jul 26 '17 at 12:49
  • $\begingroup$ Most of the answers did suggest using existing tools, but I didn't find those answers useful, there were generally answering a different (but related) question.The two most useful answers were the second half of CloudyGloudy's answer and Sasha Favorov which pointed to a preprint dealing with the same problem (which is a very stats heavy paper). $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:08
  • $\begingroup$ Well, my main point is that any answer you receive would require a relatively deep understanding of the kind of data you have and therefore your question isn't purely about statistics. Have a look at the one I cite in my question. That looks like it can be answered by any statistician. There's no need to know about RNAseq, it's essentially an issue of comparing data points. $\endgroup$ – terdon Jul 26 '17 at 13:18
  • $\begingroup$ I left an answer for the question you referenced just before I saw your post here. I'm not sure its true that any statistician could answer it. You'd need to know, for example, that RNA-seq data is negatively binomially distributed, that log TPM can in some cases be estimated by a normal, that the model is likely to need to be fit to many thousands of genes... $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:23
  • $\begingroup$ Fair enough. That might not be the best example then. But do you feel that questions that can be answered by any statistician without needing to udnerstand the details of the experimental method used should be on topic? $\endgroup$ – terdon Jul 26 '17 at 13:24
  • $\begingroup$ Sure, if someone came on and asked "How do I carry out a t-test?" or "how do I use R to fit a normal linear model" then that absolutely wouldn't be on-topic. $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:28
  • $\begingroup$ Yes, those are the types of question I am thinking of. I tried to make that clearer with my edit: "Note that if the answer requires you to understand the underlying biology, then of course it should be on topic. I am thinking of situations where the details of the dataset are irrelevant and the question is one of pure stats which could be aplicable to any field." $\endgroup$ – terdon Jul 26 '17 at 13:31
  • $\begingroup$ Okay, sure. I'm interested though why your response to this question is different to your answer to "How do we feel about questions that are essentially about text parsing?" $\endgroup$ – Ian Sudbery Jul 26 '17 at 13:40
  • $\begingroup$ Ha! Um. shuffles feet. Probably because I hate stats and enjoy text parsing. Which is an absolutely crappy excuse. But you make a very good point, it doesn't make sense to treat the two differently. $\endgroup$ – terdon Jul 26 '17 at 13:42
  • $\begingroup$ I just revisited this. I think the community is clearly in favor of stats questions, and your last comment showed up my own obvious hypocricy given how I feel about text parsing. I guess my initial position was based more on my ignorance and visceral fear of stats (I'm a biologist at heart) than on any real scientific reason. So I accepted Devon's answer since it was the most upvoted. $\endgroup$ – terdon Apr 12 at 17:06
1
$\begingroup$

Just to chip in, I'm strongly in favour of stats questions being fine given a biological context because they'll be a bioinformatics solution somewhere.

Modern bioinformatics appears to have become removed from its "statistically pure" origins but it moves cycles. At present many think (perhaps correctly at present) the volume of genomics data renders "the statistical model" irrelevant, but things change. When "the statistical model" is king, the reality is the method was shoe-horned from another area of stats/maths. The "shoe-horning" process is quite important, because your trying to place a generic solution into a biological context and the translation isn't always neat.

Example 1 For example, we just had a question asking whether "clustering analysis" can replace molecular phylogeny ... the answer is no and this question was asked about 40 years ago. However, a general statistician wouldn't be aware of the biological rational as to why, nor aware of the huge row broke out in evolution about "the correct method to build a phylogeny" and took a decade to resolve.

Example 2 A good example of "statistical methodological cycles" is machine learning (ML), which is currently fashionable, but has been on the go since the '50s. Purists describe "AI winters" (sounds like Game of Thrones, but before GoT). This is where funders pull out because the original PIs overstated their case. Without AI winters its application to bioinformatics would be much more mainstream (possibly). The reality is that many major genomics packages Sauret (whatever its called), don't implement "full-blooded" ML solution, so there's lots of ground for change.

Example 3 Finally an older but brilliant example (brilliant man) of the importance of empirical statistics specifically applied to bioinformatics was Ron Fisher's F-statistics (e.g. Fst). Fisher was a statistical purist who founded large chunks of modern statistical theory, but also applied that expertise to biological questions (just don't ask about his political views). F-statistics are used in vast numbers of bioinformatics packages and population genomics including R and I've no doubt they'll be a BioPython solution somewhere.

My advice "resist popularism" :-)

$\endgroup$
0
$\begingroup$

I would say that questions that are asking for help on statistics and nothing else should not be on topic. While statistics are an integral tool of a bioinformatician, they are not really part of bioinformatics. They're an integral tool of any quantitative science.

There is already an SE site dedicated to statistics, and I feel such questions would be better served there.

Note that I am not advocating that any mention of stats would make the question off topic, only that if the question is just asking for statistics help, it doesn't belong here.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .