Conservation of binding properties in protein models

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Drug Discovery

Background:

• The Critical Assessment of protein Structure Prediction (CASP) experiment evaluates protein structure modeling accuracy.
• Assessing the conservation of binding properties in protein models is vital for drug discovery and understanding protein function.
• X-ray crystallography provides high-resolution protein structures, but computational models are increasingly used.

Purpose of the Study:

• To evaluate how well protein binding properties are conserved when using computational models instead of experimental X-ray structures.
• To establish an accuracy threshold for reliable binding surface conservation in protein models.
• To assess the utility of protein models in predicting small molecule and protein-protein interactions.

Main Methods:

• Generated distributions of molecular probes around proteins to create 'binding fingerprints' for X-ray structures and models.
• Calculated the correlation coefficient between binding fingerprints to quantify similarity.
• Correlated binding fingerprint similarity with established CASP global accuracy measures.
• Explored ligand docking to assess small molecule binding site prediction.
• Assessed protein-protein interaction prediction through docking of models and X-ray structures to interaction partners.

Main Results:

• Binding fingerprint similarity correlates with global model accuracy, defining a threshold for meaningful binding surface conservation.
• Probe molecule clusters reliably predict binding hot spots and ligand binding sites in accurate models.
• Ensembles of models may be necessary for accurate assessment of binding pocket availability.
• Predicting protein-protein interactions is more challenging than small molecule binding, heavily dependent on local interface structure and loop conformation accuracy.

Conclusions:

• Protein structure models can reliably predict binding sites if they meet a certain accuracy threshold.
• Accurate prediction of protein-protein interactions is hindered by inaccuracies in flexible loop regions of models.
• Computational models show promise for drug discovery applications, but limitations in predicting dynamic regions need further investigation.