Related Experiment Video
Updated: Aug 6, 2026

Optimization of Synthetic Proteins: Identification of Interpositional Dependencies Indicating Structurally and/or Functionally Linked Residues
Published on: July 14, 2015
Benchmarking AI Protein Structure Predictors Reveals a Persistent Bias in Multi-State Proteins
Muhui Ye1, Yu-Hong Wang2, Maximilian Brogi1
1School of Biological Sciences, Georgia Institute of Technology, Atlanta, GA 30332, USA.
Current protein structure prediction models often default to a single dominant state, struggling with conformational ensembles. Ligands have minimal impact, while protein partners can induce state changes, limiting applications in drug discovery.
Area of Science:
- Computational Biology
- Structural Biology
- Biophysics
Background:
- Protein structure prediction tools excel at single-state accuracy.
- Their ability to model dynamic conformational ensembles and the influence of ligands or binding partners remains largely unassessed.
- Understanding these capabilities is crucial for advancing structure-based drug discovery and understanding protein function.
Purpose of the Study:
- To benchmark leading protein structure prediction models (AlphaFold3, Boltz-2, Chai-1, BioEmu) on multi-state proteins.
- To quantify state bias and sampling breadth against experimental data.
- To evaluate the impact of ligands and protein partners on conformational predictions.
Main Methods:
- Benchmarking of AlphaFold3, Boltz-2, Chai-1, and BioEmu.
- Use of four canonical multi-state proteins (PfMATE, LAO, SecA, β2AR).
- Quantification of state bias and sampling breadth against experimental reference structures.
Main Results:
- Models exhibit a strong bias towards the dominant state found in the Protein Data Bank (PDB).
- Small-molecule ligands show weak or inconsistent effects on predicted conformations.
- Large protein partners effectively drive conformational switching between states.
- Multiple sequence alignment (MSA)-based methods also display similar biases, indicating a general limitation.
Conclusions:
- Current protein structure prediction models have limitations in accurately capturing functionally relevant conformational ensembles.
- The influence of protein binding partners is more significant than small-molecule ligands in driving conformational changes.
- These findings highlight challenges for structure-guided ligand discovery and understanding protein dynamics.
More Related Videos
Related Concept Videos
Conserved Binding Sites
Binding sites are often located in large pockets, and if their location on a protein’s surface is unknown, it can be predicted using various approaches. The energetic method computationally analyses the...
Conserved Binding Sites
Binding sites are often located in large pockets, and if their location on a protein’s surface is unknown, it can be predicted using various approaches. The energetic method computationally analyses the...
Protein Organization
The primary structure of a protein is its amino acid sequence.
Protein Organization
Conservation of Protein Domains Over Different Proteins
A limited set of protein domains often duplicate and recombine during evolution. These domains can be organized in different combinations to form...
Protein Folding

