To use this software you will need a recent Phenix installation. To take advantage of the visualization tool, you will also need an installation of Coot. To use the random forest classifier, please download and decompress the file trained_rf.pkl.xz, associated with release 1.1.0 under Assets.
To run qPTxM, please supply a model (pdb or mmcif) and map (mrc or ccp4) and specify an estimated median resolution:
phenix.python qptxm.py model_file=model.pdb map_file=map.mrc d_min=3
Any modifications already present (if recognized) will be stripped off and searched for anew, and you will be shown some plots describing which possible modifications look promising and which were rejected. You can also load a copy of your model (ptms.pdb) with all proposed modifications modeled, and step through these positions with a custom script for Coot:
coot goto_ptms.py ptms.pdb
You may also like to load pruned.pdb to compare against the model with no modifications modeled. Under the menu option Custom, you will find the option to call up a panel of buttons to take you directly to each modeled modification.
Once you have decided on which proposed modifications to keep, remove any you want to reject from the file ptms.out and rerun qptxm.py with the additional argument selected_ptms=ptms.out (or whatever you may want to rename this). It will produce another model with just those modifications.
You may also find it useful to run with adjust_filters_only=True in order to test how many modifications are suggested if you adjust any of the optional parameters, like the minimum correlation coefficient of the model to the map (see below). This way is much faster but does not produce a new model and goto_ptms.py script. It also won't be able to adjust results based on an updated resolution estimate. You can also turn off plotting with plot=False or test choices of parameters on synthetic data (generated from your model but a calculated, noise-free map of the same dimensions) and see how accurately qptxm finds a randomly-modified 10% of positions in that map by setting synthetic_data=True.
qPTxM produces a file all_tested_ptms.out that contains each of the measurements it made at each of the possible modification sites. If you would like to use a machine learning method to make predictions based on these features, check out this file and the scripts in train_rf to get started. We provide a random forest classifier in the file trained_rf.pkl (compressed, distributed with release 1.1.0) and the scripts necessary to use this classifier we trained on synthetic data. The scripts write out predictions that can be passed back to qPTxM for model building and visualization.
Stojković V, Myasnikov AG, Young ID, Frost A, Fraser JS, Fujimori DG. Assessment of the nucleotide modifications in the high-resolution cryo-electron microscopy structure of the Escherichia coli 50S subunit. Nucleic Acids Research. 2020. [Full-text PDF]
Preprint posted on bioRxiv.