Marcus Povey

Structural biologists have a habit of going back to old data. Entire research projects now exist to reprocess raw datasets collected years ago with modern tools, and they keep finding things the first analysis missed. That is a good reason to keep the raw data, but it is also a problem, because it is rarely anyone’s funded job to keep it.

This week, I attended the EOSC Symposium, wearing multiple hats, and representing both Instruct, in my capacity as head of the Instruct Digital Services group, and as technical officer on the Life Sciences Connect EOSC node. We were there to build EOSC, an initiative to build out the infrastructure to conduct open science within Europe, technical, policy, people.

There was networking, there were keynotes, and there was even dancing (although less said about that the better).

One of the recurring themes throughout is the need for access to well curated trustworthy data, and I wholeheartedly agree. This is doubly important given the advent and use of AI, making once-hard problems tractable, provided the data is there. The field of Structural Biology has led the way amongst the life sciences, and breakthroughs like AlphaFold have accelerated drug discovery research.

I was privileged to be invited to speak at the event, and I was glad to be able to talk about some of my work at Instruct, building the provenance chain from initial scientific proposal, through to raw data collection, processing and final output. It was here that I believe I was able to highlight a gap in a number of assumptions within the EOSC when it comes to the importance of data, namely, it is not just the final result output that is important, it is also the raw data and provenance chain that led up to that output.

Within the field of Structural Biology, there are entire research projects engaging in “data archaeology”; reprocessing old raw data sets with more modern tools and getting improved results and new discoveries. AI models turn this process up to 11, except now you also need to preserve metadata alongside the raw data to include things like machine settings, make and model, details around the sample, details around the preparation and protein production, etc.

The effect, when we think about FAIR data, is that in terms of long term archival, we’re probably adding a couple of orders of magnitude to the volume of data that needs to be stored.

Right now, final outputs are largely handled by the repositories (PDB, BMRB, EMDB, and so on). Some raw data is archived too, for example cryo-EM images in EMPIAR, but coverage is partial, and there is rarely a systematic record linking a raw dataset to the sample, the instrument settings, and the processing that produced the final model. For my part, this is some of what we are working on at Instruct, building a provenance chain associating the original science to the final output, but with details of every step along the way. However, for the most part, the raw data is still held at our facilities, for which there is no uniform capacity for long term preservation, and even well equipped facilities can struggle.

I have been working with EOSC projects (EOSC Beyond) and other EOSC nodes (EOSC Poland, EGI and EUDAT) to address part of this problem. Namely, it is now possible within ARIA to broker storage from a number of providers at collection time, giving a researcher a “dropbox” to store their data for longer term access. However, this is currently only a pilot (albeit a successful one).

So, it was with this in mind that I was banging the drum for addressing long term data archival at the European level as a critical need for our community (and not just Structural Biology, the Photon and Neutron community have a very similar problem).

The challenges are great, as are the data volumes, and at some point somebody needs to pay. Archival is also not just a matter of building a central data centre. It needs redundancy across geography and jurisdictions, and in an uncertain geopolitical climate that resilience has real funding implications.

It is, however, a gap I believe that EOSC could really prove its worth by addressing.

---

I write this in my rented apartment in Florence, the sounds drifting in through the open window from the narrow street outside. I am in a “local” part of the city, far enough away from the maddening tourist hordes (of which, I guess, I count amongst their number), but close enough to hear the bells of the cathedral on the wind.

I have just come back from a drink at the cafe on the corner, sat with my book and a glass of wine, watching the world go by in the late October sun, pretending to be a local resident, rather than a rootless nomad. Soon, I will leave, back to my regular haunts, via the Hive city of London.

I will be sad to leave, “Home” feels like a memory.

The views in this post are my own and do not represent Instruct-ERIC or the EOSC Life Sciences Connect node.