Bioinformatics Zen

I think simple is better. Statistics says so too. Statistics says that you’ll probably read the first two paragraphs of this post, look at the pictures then go elsewhere. So I’d simply better get to my point. In terms of attention spans, computer code and (statistical) explanations, and possibly everything in general, I think it’s always better to favour simplicity over complexity.

Coding
Creating computer code can get pretty tricky. When things stop being simple is when bugs start appearing. You could call software bugs “unexpected complexity”, things start happening you don’t want to. Everything is lot easier to debug if it’s simple. For example, using an existing library rather than write new code. Simple, well-commented code also makes it easier for other people to understand what’s going on. Simplicity doesn’t mean short though, collapsing several lines into a single one, whilst ingenious, will make the code more difficult to understand in the future.

Presenting
People have limited attention spans. Think about what you want the audience to really remember from your talk, focus on this. Deliver the message simply. Drop the bullet points / clip art / weird backgrounds. Get your content inside the audiences head effectively as possible. If they want any more information they can ask you, or read your paper.

Writing
Again the reader has a limited attention span. Background is important no one wants to wade through 2/5/10 pages of blah blah blah before getting to the point. Who ever is giving up their time to read a paper/report/dissertation is doing so to understand the point of the work. So give it to them. Revise the text as many times as you can take, and cut as much as possible. Use the active voice. Keep the thread vigorous and punchy. No one wants to read bland text, and it’s pretty tedious to write as well.

Statistics
The principle of parsimony prefers the simplest explanation over the more complicated. Why use 20 variables when 5 will do just as well. Of course if you need 20, you need 20. But over fitting a model will usually result in less predictive power.