Modeling Alternative Conformational States in CASP16.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• The CASP16 Ensemble Prediction experiment evaluated computational methods for modeling molecules in multiple conformational states.
• This study focused on multi-state targets with experimentally determined structures, excluding domain-linker-domain targets assessed by statistical models.

Purpose of the Study:

• To assess the progress in computational modeling of proteins, nucleic acids, and their complexes in various conformational states.
• To identify successes and persistent challenges in predicting multi-state molecular structures.

Main Methods:

• Ten multi-state targets were released as community challenges, including ligand-induced changes, protein-DNA complexes, and RNA structures in different oligomeric states.
• Predictions were evaluated by direct comparison to experimental coordinates.
• Successful approaches often involved generating multiple AlphaFold2 models with enhanced sampling and sequence alignments, followed by quality-based selection.

Main Results:

• Reasonably accurate models (TM-score > 0.75) were achieved for five targets, particularly one protein-ligand complex (T1214) using an apo structure as a template.
• Predictors generally failed to capture key distinguishing structural details between states, except for the T1214 case.
• Accuracy was significantly lower than for single-state targets in previous CASP experiments.
• Predictions for protein-DNA complexes, RNA targets, and multiple RNA oligomeric states generally fell short (TM-score < 0.75).

Conclusions:

• While progress has been made, accurately modeling conformational ensembles, especially for RNA and large multimeric assemblies, remains a significant challenge.
• AlphaFold2 and AlphaFold3 showed promise, but individual group performance varied, sometimes outperforming servers.
• Multi-state prediction represents an important frontier in structural biology research.