After 10+ years working with #metabolomics mass-spectrometry data, you would think I wouldn't be surprised any more.

However, I'm surprised every time when a method or tool uses non-log-transformed abundances into a standard t-test or another semi-normal distribution assuming statistical framework. 🤦‍♂️

Seriously, we need more statistically minded people in #metabolomics / #lipidomics.

https://rmflight.github.io/posts/2021-04-09-proportional-error-in-mass-spectrometry/

Deciphering Life: One Bit at a Time - Proportional Error in Mass Spectrometry

@rmflight The problem is everyone wants to do metabolomics and it’s made so much more accessible that anyone with a mass spec can do it. That means those doing metabolomics maybe just chemists or even pathologist or *insert any other expertise* who don’t see “metabolome” or it’s dynamic scale. Common assumption is ion intensity = concentration.

@drupad Yep! And of course we haven't had the yelling from the rooftops like we did in early microarray days and I think very quickly w/ RNA-seq that you can't use non-transformed intensities.

What's really disturbing, is that a well known company in this space knows internally that they should use log-transformed values, but their customer facing product for investigation has the "log-transform" radio button un-checked by default. 🤷‍♂️

@rmflight I might know a similar company too and have a VERY interesting discussions with one of the team members at ASMS last year. Alas, change is slower than expansion of the community for this omics. And inventions in mass spec space has made it worse. The rate of new data generation is faster than the acceptance of better ways to process, handle and interpret the data.
@rmflight just checking… you’re saying that there are scientists doing statistical tests without checking the distribution of the data first? 😬
@rmflight cld rely have just stopped at “people” at the end, there.