Comparing atom-based with residue-based descriptors in predicting binding site similarity: do backbone atoms matter?

Area of Science:

• Structural bioinformatics
• Computational chemistry
• Drug discovery

Background:

• Predicting polypharmacology requires accurate protein binding site similarity detection.
• Current methods may not capture subtle structural nuances relevant for identifying cross-reactivity.

Purpose of the Study:

• To investigate the optimal level of detail in protein 3D representations for detecting binding site similarity.
• To assess the performance of an enhanced site comparison algorithm (SiteAlign2) for polypharmacology prediction.

Main Methods:

• Modification of the SiteAlign program to incorporate atom-based descriptors accounting for solvent exposure.
• Benchmarking SiteAlign2 against the original SiteAlign using protein datasets.
• Evaluation of performance based on sequence and structural similarity metrics.

Main Results:

• SiteAlign2 did not show global performance improvement over SiteAlign.
• SiteAlign2 demonstrated enhanced success in identifying similarity between non-homologous proteins when backbone atoms were critical for ligand binding.
• The new method is more sensitive to atomic coordinate precision.

Conclusions:

• SiteAlign remains effective for comparing binding sites of homologous proteins.
• SiteAlign2 offers improved insights into physical binding site similarity for non-homologous proteins.
• The choice of method depends on the degree of protein relatedness and the desired level of detail.