Comment by xkcd-sucks | Hacker News Reader

Comment by xkcd-sucks | Hacker News Reader

Parent Full thread

All these points are correct.

But, to be fair, the state of the art in biology is still full of Magic and other nonreproducible effects.

Computery people get annoyed if a compilation takes 10 minutes and fails with a cryptic error message. Compare this to an experiment where (for example) you spend a full day harvesting fetal rat hippocampus cells, cultivate them for weeks, and then finally run your experiment which "fails". Did you fail the rat prep? It worked for the wizard postdoc but she has been dissecting tiny little hippocampi for years -- maybe the cells aren't what you think they are. Or maybe, due to some unknown factor (temperature? feeding schedule? co2? mycoplasma? ...), the cells are "unhappy" for... some reason . Etc. More of the same, another month, another year, another replication which may or may not be sufficient to quit the experiment altogether. The systems are not understood well enough to debug; the systems which are easily debuggable are not relevant to the real world. But, it is not difficult to do at home what one does in an academic lab. (corporate labs have a bit more money which shortens the iteration loop a bit but doesn't make it less cryptic)

Anecdotally, I got a life science PhD and went into a softwarey career because the iteration loop of biology just plain sucks. When I was in grad school, I had free reign to do absolutely whatever with no oversight (if the bills looked normal) -- and did, for a while. All those cool ideas I had, I am content to let other people have and realize them. I am content to read their papers and to hire them.

>But, to be fair, the state of the art in biology is still full of Magic and other nonreproducible effects

No, it's not. It's full of things that have week long design cycles (instead of a 200 ms compile loop) and are a pain, and lots of grad students that have either never heard of design of experiments, or know about it and are overeager to apply it. A friend worked at BP biofuels. The team kept waving off irreproducibility as 'biology being strange'. For months she kept saying "hey guys our feedstocks aren't defined". Then finally after a half a year they caved and let her do her experiments, and sure enough there was a limiting nutrient in the feedstock that correlated with the fluctuations. Then a few months later the supreme court ruled against BP and the biofuel project got cancelled.

I had a similar experience, and switching model organisms took my design cycle from 2 weeks to two days, and got huge results out of it. We then put the results back in the original organism and recapitulated the improvement. Then my project got cancelled.

In the end though, it seems I'm better at software and it pays better.

> I had a similar experience, and switching model organisms took my design cycle from 2 weeks to two days, and got huge results out of it. We then put the results back in the original organism and recapitulated the improvement. Then my project got cancelled.

Very cool. If you care to elaborate on this: what were the two organisms, what made the second amenable to much faster design cycles, and how did you know that the results would likely carry over to the original organism?

Slow organism was synechococcus, fast organism was E. Coli, original results were highly likely to carry over since it was a fundamental alteration to the enzyme function.

https://pubmed.ncbi.nlm.nih.gov/23819621/

If you use the right tools and live in the right place you can get an e coli design cycle down to 48 hours. In San Diego, i could design primers in the afternoon, have them on my desk in the morning, do a pcr/Gibson assembly that was done by lunch, pick colonies in the evening, miniprep and retransform early the next day and have phenotypes soon after.

It's also robust. I trained an intern with only classroom experience in molecular biology and he made 50 mutations and biochemically tested half over the course of 10 weeks.

https://pubmed.ncbi.nlm.nih.gov/24934472/

Thanks for sharing your views. Do you see the iteration loop in bio remaining infuriatingly slow, or is this something that improves ever year as tools get better / machines get faster / more biotech microservices arise to remove tedium?

What do you do now? Any links to your old research? Sounds quite interesting. If you have the chance, get into computational biology, it is getting quite popular right now.

Reply on news.ycombinator.com