Green Intentions November - Gin tonic, data and invisible waste: preventing unsustainable addictions by Estel Collado Camps

It was a gorgeous, sunny late summer afternoon. Gin tonic in hand, I was chatting with two other wedding guests. We were an engineer and two scientists camouflaged under a nice suit and two summer dresses. And it happened again. The bubbles, the happy occasion and the sun were not enough to keep us away from a serious discussion. About data - and the future of it.

Data is not only problematic when it is owned by Facebook. The amount of data we produce is growing so much that storage is a concern. Remember: data centres also make tape copies of what is stored in their servers. And producing data is also costly. While the molecular storage field is trying to use DNA or fluorophores as more efficient ways to store data, most scientists are still stimulating each other to produce as much data as they can. After smoking, alcohol and red meat… is data our new addiction? Maybe we should talk about it in lab meetings, instead of at cocktail parties.

Let’s first take a step back. Are you sure that Meten is weten? I would like you to consider the practical link between data and knowledge. Imagine I give you a bunch of datapoints without telling you where they come from. The only knowledge you gain is not to trust me. Which might be useful for you, but… you know. Any resources that went into the production and storage of these data could be used in a much better way. That is not very sustainable.

Sadly, this happens frequently in research. Sample sizes we can’t draw any conclusions from problems finding or reproducing what the previous student did, results that will never see the light because they’re negative. Results stored in an obscure hard-drive somewhere, or behind a paywall. If data is supposed to become knowledge, it needs to be seen. Interpreted, combined, used. While we try to reduce plastic, fly less and eat less meat, we neglect the importance of thorough consideration when it comes to data production, storage, sharing and reproducibility.

My own examples make me feel uncomfortable, but I’ll share them. If we all put our uncomfortable results out there, others could focus on doing something better, right? To start with, I thought the FAIR criteria were not relevant for me. Three years into my PhD, I bumped into a database about cell-penetrating peptides. Ergo: there are more ways than you think to make your data Findable and Reusable. Maybe the FAIRshake tool offers some more.

Secondly, I spent four years of my PhD refusing to learn (the basics of) R. Before that, anything I’d done to my raw data was useless without access to GraphPad. In the end, a lockdown and the perseverance of a good friend made me become an R markdown fan. If we all used as much open software as possible, we would help the Interoperability principle. Bye-bye stress about the renewal of the institutional GraphPad license.

Third, I thought power calculations were just a requirement for animal study protocols. Actually, they can trigger us to think differently about experimental planning. So, which effects and how much variation do you expect, and what will really be biologically relevant? Needless to say: the perverse association we made between statistically significant and relevant is quite twisted. Constraints like costs and number of patients available are unavoidable issues. Then we should acknowledge them. We should be seeing pilots as a way to know the SD for the next power calculation. Instead, we are tacitly tweaking power calculations to get to the desired sample size.

Constraints should also be seen as a stimulus to sharing and reusing data. Here’s a publication showing new things by looking at already existing data. Also, results from control samples could be reused. But for that, you need to make sure that new and old data can be combined. Which brings me back to the wedding conversation. Is it so, as our engineer put it, that we should collect all the data we can, even if we’re not sure what we’ll use it for?

The idea behind this affirmation is that questions will arise, and otherwise AI will find hidden patterns that we can’t conceive. To me, this sounds again like an unsustainable addiction. Remember where I started, with that mysterious bunch of datapoints. Now imagine we invoke some AI power which can even infer what sort of data this is. Was it still worth it to create these data a priori and store them for a few years?

Seeing how fast techniques evolve, probably we can produce the same data better, faster and cheaper in the future. If we produce data when we need it, we save storage costs. However, for rare diseases it could be great to accumulate data: we’re all mortal, and patients do not come back. But can samples from 10 years ago be combined with new ones? Our enthusiastic engineer said that AI would help with this as well. To me, one thing is clear: without good data management, combining past and future datasets won’t work. Which ties back to the FAIR principles.

Also, the questions of the future may be related to phenomena that are not here yet. I’d rather make the choices I can make now. What can YOU do to make the most out of your scientific efforts? In the meantime, please save server pipette tips and server space for the next crisis. Drink less data but of better quality. Just like with good wine: cherish every bit and share great research with the world.

The Green Lab Initiative is a new collaboration of green enthusiasts at the Radboudumc who want to encourage people to go greener in the lab. With our green intentions of 2021, we try to inspire you every month to take small steps. Find us on Twitter or LinkedIn!

Blog by Estel Collado Camps