atfrank
commented
3 years ago Either way, for the manuscript, Metallo should only be using the RNA coordinates to generate dummy atoms. For the use case we have in mind, the user only uploads RNA coordinates.
On Fri, Sep 17, 2021 at 2:51 PM Lichirui Zhang @.***> wrote:
Hi Jingru,
I wonder if the classifier was trained by the features in train/data/fingerprint?
I found that the features generated by the current Metallo is inconsistent with the feature in train/data/fingerprint. I am using the script in the "sh" folder. For example, 1d4r, the current Metallo generates 535 dummy atoms, however, train/data/fingerprint/1d4r.txt only contains 25 dummy atoms.
I suspect the features in train/data/fingerprint were generated by an earlier version of Metallo, which was committed by me on Feb 7, 2020, [ https://github.com/atfrank/AtomicFeaturizer/tree/5e026fa22b26451b5f5fca1a9fde35e369e2ad6f]. In that version, Metallo still retains crystal water, and uses them as a constraint to generate dummy atoms (I think the current version also retains crystal water, but the water distance upper bound constraint is relaxed to 10000 Å). The related line is Line 315 in the current and the earlier metallo.cpp.
Thanks, Lichirui
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/atfrank/Metallo/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABM6IQB4VOZ2WW32ZFMKTYDUCOE4LANCNFSM5EIEOA7Q .
-- Aaron T. Frank (he/him) Assistant Professor of Biophysics Board of Reviewing Editors, eLife Advisory Editorial Board, Biophysical Chemistry Early Career Board (ECB), Journal of Chemical Information and Modeling
University of Michigan 3000 Chemistry, 930 N. University Ave. Ann Arbor, MI, 48109 | (734) 615-2053
https://sites.lsa.umich.edu/frank-lab/ https://twitter.com/aFrankLab
Hi Jingru,
I wonder if the classifier was trained using the features in train/data/fingerprint?
I found that the features generated by the current Metallo is inconsistent with the feature in train/data/fingerprint. I am using the script in the "sh" folder. For example, 1d4r, the current Metallo generates 535 dummy atoms, however, train/data/fingerprint/1d4r.txt only contains 25 dummy atoms.
I suspect the features in train/data/fingerprint were generated by an earlier version of Metallo, which was committed by me on Feb 7, 2020, [https://github.com/atfrank/AtomicFeaturizer/tree/5e026fa22b26451b5f5fca1a9fde35e369e2ad6f]. In that version, Metallo still retains crystal water, and uses them as a constraint to generate dummy atoms (I think the current version also retains crystal water, but the water distance upper bound constraint is relaxed to 10000 Å). The related line is Line 315 in the current and the earlier metallo.cpp.
Thanks, Lichirui