RoseTTAFold VS ColabFold

I wonder what the license for RoseTTAFold is. On github you have:

https://github.com/RosettaCommons/RoseTTAFold/blob/main/LICE...

But there's also:

https://files.ipd.uw.edu/pub/RoseTTAFold/Rosetta-DL_LICENSE....

Which is it?

so far the options are: - https://github.com/RosettaCommons/RoseTTAFold/tree/main/example/complex_modeling : look for make_joint_MSA_bacterial.py. For eukaryotes you're out of luck. - https://zhanggroup.org/cpxDeepMSA/ haven't tested it yet. waiting for results. - if you already have MSAs for individual proteins you can try to map some with taxidID (if they have it provided in the Uniref header). Still looking for some more elaborate methods using phylogeny.

Code for RoseTTAFold: Accurate prediction of protein structures and interactions using a 3-track network

https://github.com/RosettaCommons/RoseTTAFold

In the science article, they use it to predict theoretical structures of heterotrimers (https://www.science.org/doi/10.1126/science.abj8754) , however upon looking in their github (https://github.com/RosettaCommons/RoseTTAFold) they only seem to have prodcedure for predicting complexes of heterodimers. I've successfully done the procedure to predict heterodimers of proteins of interest to our lab, but cannot figure out how to predict trimers from the github

Baker Lab (https://www.bakerlab.org/) is cutting edge in that area and also has a protein prediction open source package (https://github.com/RosettaCommons/RoseTTAFold)

RoseTTAfold paper , GitHub

and RoseTTAFold is here: https://github.com/RosettaCommons/RoseTTAFold

Thanks for the info. Any opinions on ColabFold?

I'm on the software & infrastructure side of things, so perhaps don't have the depth of the answer you're looking for, and would be happy to be corrected. But as far as I understand from the scientists on our team is that "it depends". Both can get results that are remarkably close to one another, but both can also disagree, so it's hard to say what's "better" - it depends on what you consider the ground truth for your experiments, and a number of other factors such as the length of the sequences you're working with. Both are able to predict structure from sequence, recover sequence from structure, as well as "hallucinate" structure and sequence, so it can be helpful to use both to validate your results and increase your confidence. I assume you've seen the ColabFold notebooks, so it should be easy to compare them against each other for your use case.

a word of warning, the requirements are pretty high, if the sequences are not too long try one of the notebooks: https://github.com/sokrypton/ColabFold

In the ColabFold GitHub README there is a description of differences between ColabFold, DeepMind's version, and others. Apparently, the usage of templates and mmseqs2 was not available in the default DeepMind's implementation and was added in ColabFold. https://github.com/sokrypton/ColabFold