Method developed by Stephanie Kim (kim.6088@osu.edu) and Justin Seffernick (seffernick.9@osu.edu).
Code and documentation by Justin Seffernick.
The PI was Steffen Lindert (lindert.1@osu.edu).
Last updated August 2021 by Justin Seffernick.
Application resides at Rosetta/main/source/src/apps/public/analysis/ResidueDisorder.cc.
There is a demo for disorder prediction using ab initio structures at Rosetta/demos/public/ResidueDisorder.
Note that to run ResidueDisorder for intrinsic disorder prediction from sequence, structure prediction must be performed prior. This can be done using Rosetta ab initio, RoseTTAFold, or AlphaFold. Structures must be relaxed in Rosetta prior to input. To run ResidueDisorder for measuring disorder from structure, each input structure must also be relaxed.
Kim, S.S.; Seffernick, J.T.; Lindert,S., Accurately Predicting Disordered Regions of Proteins Using Rosetta ResidueDisorder Application. J. Phys. Chem. B. 2018;122,3029-30930.
Seffernick, J.T.; Ren, H.; Kim, S.S.; Lindert, S., Measuring Intrinsic Disorder and Tracking Conformational Transitions Using Rosetta ResidueDisorder. J. Phys. Chem. B. 2019;123 (33), 7103–7112.
Rosetta ResidueDisorder is a method to predict intrinsically disordered regions of proteins from the primary sequence. While proteins are often represented as well-folded, rigid structures, many proteins contain regions that are unlikely to form stable structures. The ResidueDisorder app (to be used after running structure prediction and relax prior) can be used to predict order/disorder at a residue-resolved level. Structure prediction can be performed using Rosetta ab initio, RoseTTAFold, or AlphaFold.
Additionally, ResidueDisorder can measure disorder of different conformations of the same system directly from the coordinates from an MD simulation for example (or any ensemble of structures). Using this methodology, ResidueDisorder can also predict folding/unfolding events from a trajectory based on changes in disorder.
To predict intrinsic disorder from sequence, relaxed models are first generated (Rosetta ab initio, RoseTTAFold, or AlphaFold, performed prior to the input into the ResidueDisorder application), the average score for each residue is calculated. To measure disorder directly from a single structure, the structure is relaxed and residue scores are calculated. Both methods proceed with the following steps.
For each residue, the average score is smoothed over a window size (WS), resulting in the residue-resolved order score (OS). For example, when calculating OS for residue N with WS of 5, the OS is defined as the average residue score of N and the 10 (WS*2) adjacent residues (N-5, N-4, N-3, N-2, N-1, N, N+1, N+2, N+3, N+4, N+5). If N is within 5 residues of either the N- or C- terminus, only residues that exist were included. For example, the order score of residue 2 would result in averaging the scores of residues 1, 2, 3, 4, 5, 6, and 7. Next, each residue is predicted as either ordered or disordered, depending on whether the order score is below or above the cutoff value (CV), respectively.
If the protein is more than terminal percentage cutoff (TPC) percent disordered (percent of residues predicted as disordered) or if AlphaFold was used for modeling, the algorithm is complete. However, if the protein is less than TPC percent disordered, a terminal residue optimization is applied. The terminal optimization step changes the cutoff for residues within terminal size (TS) percent of either terminal residue. For these residues, a linear cutoff line is applied such that the N- and C- terminal residues have a cutoff of terminal cutoff value (TCV) and the TPC percent residues have a cutoff of CV (along with all other inner residues) with a linear extrapolation. Again, each residue is predicted as ordered or disordered depending on whether the order score is below or above the new cutoff.
The algorithm contains parameters for modeling using the Talaris2014 and REF2015 scoring functions, as shown in the table below. AlphaFold with REF2015 is recommended for best prediction results.
| Model generation method | Rosetta scoring function | WS | CV | TPC | TS | TCV | Flags to run |
|---|---|---|---|---|---|---|---|
| Rosetta ab initio or RoseTTAFold | Talaris2014 | 5 | -1.0 | 60% | 13% | -0.3 | -restore_talaris_behavior |
| Rosetta ab initio or RoseTTAFold | REF2015 | 10 | -1.5 | 40% | 34% | -0.8 | No additional |
| AlphaFold | Talaris2014 | 5 | -1.2 | NA | NA | NA | -restore_talaris_behavior and -AF |
| AlphaFold | REF2015 | 10 | -1.8 | NA | NA | NA | -AF |
If event prediction is performed (from frames of an MD trajectory), the difference in the average disorder for 25 steps before that timepoint and 25 steps after that timepoint is calculated. A cutoff line for the absolute value of the difference is instituted, above which an event is defined, representing a significant change in disorder at that timepoint. This cutoff line is defined as 60% of the largest difference for the system. If the disorder increases during the event, it is defined as an unfolding event and if the disorder decreases during an event, it is defined as a folding event. If events are detected in adjacent timesteps, they are combined to form a single event in the center of the range.
Before using the ResidueDisorder application to predict intrinsic disorder from sequence, structures models must be generated using Rosetta ab initio folding (100 recommended), RoseTTAFold (at least one recommended), or AlphaFold (5 recommended). The only input needed to run the application is the generated structures (using the -in:file:l flag is recommended).
Before using the ResidueDisorder application to measure disorder from structure, each structure must be pre-relaxed. The only input needed to run the application is the relaxed generated structures (using the -in:file:l flag is recommended). If event detection is performed, it is crucial that these structures be input in the correct order.
For predicting intrinsic disorder from sequence (default behavior of application), ResidueDisorder can be run with any number of folded structures, but recommended numbers are shown above. To use the AlphaFold parameters, include the -AF flag. The default output is ResidueDisorder_default.out, but the output file can be specified using -out:file:o.
For measuring disorder from structure, the flag -measure_disorder_from_structure must be given. This will treat every input structure separately rather than averaging the ensemble. The default prefix for the output files (one for each structure input) is ResidueDisorder_default_, but the prefix can be specified with -out:prefix.
To predict events, make sure the structures are input in the correct order. The flags -measure_disorder_from_structure and -predict_events must be given. To specify the prefix for the output of each structure, use -out:prefix. To specify the output file containing the events, use -out:file:o (ResidueDisorder_events_default.out is the default output).
Also, make sure that all input structures are of the same sequence (and length, must contain at least 10 residues) and contain only a single chain.
For measuring disorder and predicting events, the current implementation is calibrated to be used with Talaris2014, so the flag restore_talaris_behavior must be given. However, for prediction of intrinsic disorder, REF2015 can be used (see table above in Algorithm section for specific flags for each implementation).
Currently, Rosetta ResidueDisorder can only predict order for monomers. Input structures must be of a single chain.
ResidueDisorder will output (on screen) the calculated per-residue scores for each pose and the average per-residue scores (over all poses if predicting intrinsic disorder from sequence). For the initial prediction, ResidueDisorder will output the order score and prediction (order or disorder) for each residue. If terminal optimization is necessary (less that TPC% disordered), ResidueDisorder will output the terminal optimization order scores, predictions, and final cutoffs.
If predicting intrinsic disorder from sequence (default behavior), ResidueDisorder outputs the final predictions in tabular format to an output file (name is ResidueDisorder_default.out by default, but output file name can be specified using the -out:file:o flag). The tabulated final results contain the prediction, order score, and raw average residue score.
If measuring disorder from structure (-measure_disorder_from_structure), ResidueDisorder outputs the final predictions in tabular format to an output file for each input structure (prefix is ResidueDisorder_default_ by default, but file prefix can be specified using the -out:prefix flag). The tabulated final results contain the prediction, order score, and raw average residue score.
If event detection is performed (-measure_disorder_from_structure and -predict_events), in addition to the output results for each structure (see above), an additional output file is created containing the detected events (ResidueDisorder_events_default by default, but can be specified using the -out:file:o flag).
Since the first release, the "measure disorder from structure" and "predict events" capabilities have been added as well as parameters for predicting disorder using AlphaFold structures.