17th Community Wide Experiment on the
Critical Assessment of Techniques for Protein Structure Prediction
Target: E2459
Target: E2459
Type: All-group target
Entry Date: 2026-07-30
Server Expiration Date: 2026-08-06
Human Expiration Date: 2026-08-30

Protein: GABARAP
Organism: Homo sapiens
Residues: 117
Method: EM
Additional Information: This is a hidden-states FRET ensemble target. The goal is to model the one, two, three, or four dominant conformational states for the target and assess these against FRET distance data provided by Claus Seidel (Heinrich Heine University Düsseldorf). The target is the small monomeric Homo sapiens GABARAP (UNIPROT ID: O95166) in its soluble form (117 aa, 13.92 kDa). GABARAP is part of the autophagy machinery. GABARAP's localization shifts from the cytosol to the membrane by post-translational modification with a phospholipid membrane anchor.

For soluble GABARAP, 15 variants with FRET pairs at distinct positions have been used to generate ground truth data for 15 interdye distances. Several conformer-specific sets have been found. In this experiment, predictors will submit multi-state models. Note that each conformer-specific set consists of 15 interdye distance data that have been captured for assessing the proposed states. For your information, we applied fluorescence correlation spectroscopy and found multiple FRET relaxation times n-1=3, where n is the number of kinetically relevant conformer species. In this case, n=4 kinetically relevant conformer species were resolved by these FRET data. The relaxation times range from 2 and 20 to 200 µs. Thus, it is obvious that the energetic barriers between the conformers differ significantly. On the one hand, it is difficult for classical methods to resolve these short-lived states, and on the other hand, it is difficult to simulate them because the relaxation times are quite long for simulations of dynamics.

Please provide model "ensembles" of conformational states for the provided sequence. Predictors may model the conformational ensemble as one, two, three, or up to four distinct states with potential substates, submitting up to 100 models for each state. Models 1–100 should represent state v1, models 101–200 state v2, models 201–300 state v3, and models 301–400 state v4. Note, for example, that the best ensemble prediction may be only 1 state (v1), with no models submitted for v2, v3 or v4; or two states (v1 and v2), with no models submitted for v3 or v4. We request that estimates of the relative populations of each state (a decimal value between 0 and 1.0, summing to 1.0) and a comment regarding why you chose one (or more) states are provided in the populations.txt file. This file should be tarred together with the coordinate files and submitted through a dedicated gateway https://predictioncenter.org/casp17/predictions_submission_ENSMBL.cgi, which provides further instructions on preparing submissions. The results of the format checks can be tracked through the notification emails. If you need to fix your models and resubmit, please do this according to the original submission procedure.

Sequence: (Plain text version)
Template: (Plain text version)
Protein Structure Prediction Center
Sponsored by the US National Institute of General Medical Sciences (NIH/NIGMS)
Please address any questions or queries to:
© 2007-2026, University of California, Davis
Terms of Use