Thread by @SjorsScheres

As this thread is gathering all kinds of perspectives, below are mine, including on a bit of history. But before that, just to say that I very much value the hard work by all the people who make #cryoEM image processing better for everyone, both in academia and in industry.

https://twitter.com/biochem_fan/status/1666751772084535297

In 2012, I introduced a regularised likelihood approach for #cryoEM structure determination, which formed the basis for #RELION. This approach proved to be hugely successful and is now the basis of many software packages, including also #cryoSPARC.
sciencedirect.com/science/articl…

When #RELION started to replace #EMAN and #SPIDER, and before #cryoSPARC, I expressed in multiple talks that the #cryoEM field needed more diversity in software, as a monoculture is vulnerable. I used the picture below in my talk at the 2014 NRAMM course.

In 2017, @a_punjani and @marcusabrubaker made a brilliant contribution to the field by the introduction of SGD for initial model generation. This was the beginning of #cryoSPARC.

Unaware that this approach would be patented (something that was not disclosed in the @naturemethods paper) we implemented it also in #RELION. This was easy, as #cryoSPARC’s algorithm built on #RELION's regularised likelihood approach.

When we found out about the patent, we asked our legal at @The_MRC what to do. They got very stressed, and many meetings with them later, I removed all SGD algorithms from #RELION. I now stay far away from any algorithm in #cryoSPARC.

It does not work like that in the other direction. When we published our @biorxivpreprint on aberration correction in #RELION, these algorithms were implemented straight away into #cryoSPARC (faster than @IUCrJ published our paper).
biorxiv.org/content/10.110…

But @structurabio using our algorithms is completely fine! This is the intention of open-source software. We are happy with the impact of our development, and glad to see it being useful for others too. This is how science progresses: we all build on each other’s advances.

But if one views the #cryoEM field as a competition for users, perhaps as implied from @biochem_fan’s initial tweet, then the asymmetry of the direction of algorithm development and uptake in the respective softwares will inevitably lead to the trend just observed.

There are positives here! @structurabio is extracting money from industry and pouring it into the #cryoEM field through user-friendly, fast software that would be difficult to make in an academic setting. Academic users are profiting, as they can (for now) use it for free.

But there are a few developments that worry me. In the commercial sector, small companies eventually tend to end up in large ones. For life science, the end-point often tends to be @thermofisher.

Already, you can buy a Krios from them with a #cryoSPARC package included. How will that look like in 5 years time? Will #cryoSPARC still be free for academia? Would @thermofisher monetise on a monopoly?

Another worry is that it is not clear how cryoSPARC exactly works. Is the field comfortable with black-box software?

We just encountered a practical example at @MRC_LMB, where our PB-sized disks are failing due to unknown causes. Disk access patterns in #cryoSPARC have been flagged as a suspect by our scientific computing facility, but the closed-source makes it hard to get to the bottom.

So, there are positives and negatives. In some cases #RELION works better, in other cases #cryoSPARC gives better results. The hard work by dedicated developers from both teams is extremely valuable, and perhaps even better software is coming from new players in the future.

We will soon report a new feature in #RELION that will again move the goal posts. Perhaps it will also become available in #cryoSPARC? In that sense, competition is working as it should. As long as more than 1 software survives, this should continue and science will win. 🤗