Recognizing protein-ligand binding sites

In this paper, Roy and Zhang describe COFACTOR, which predicts protein-ligand binding sites. COFACTOR is a comparative and hierarchical approach that uses structure modeling and a global-and-local similarity search. This method outperformed all other methods in the CASP9 competition, thus highlighting the importance of their approach.

The authors first carry out structure modeling based on the I-TASSER algorithm. Following this, a global similarity search is performed to identify template proteins with bound ligands using TM-align. During the local similarity search, sequence and geometrical information is considered in a step-wise manner: first, considering evolutionary information by position-specific iterative basic local alignment search tool (PSI-BLAST) and the Jensen-Shannon divergence score, and, secondly, by structurally aligning the candidate binding-site motif to the template motif using Needleman-Wunsch dynamic programming. Ligand conformation is ultimately refined using a quick Metropolis Monte-Carlo simulation. 

Evaluation of this method showed that COFACTOR accurately identifies 65%-69% of ligand-binding pockets and interacting residues with a Matthews correlation coefficient (MCC) of 0.55-0.58. Furthermore, it was shown to perform better that all other methods in the CASP9 competition. The authors argue that its success resides on the combination of both local and global structural alignment. 

The future of protein modeling: integrative modeling

In this paper, Russel et al. set up the grounds for the next generation of protein modeling platforms that combine computational modeling with experimental information from various sources. This will permit us to re-assess and refine published structural models as new structural information becomes available and suggest future experiments that validate the refined model in a feed-back loop.

In the integrative modeling platform, models are encoded as a collection of particles. Each particle can be used to create atomic, coarse-grained, or hierarchical representations. Models are then evaluated by a scoring function composed of terms called restraints. Each restraint corresponds to one particular experiment and measures how well a model agrees with the structural information derived from that experiment. The IMP’s scoring function includes restraints for small-angle X-ray scattering (SAXS), mass spectrometry, electron microscopy, nuclear magnetic resonance (NMR), Chemistry at Harvard Macromolecular Mechanics (CHARMM) force field, statistical potentials, alignment with related structures and 5C data. 

IMP also integrates different sampling algorithms that will allow users to refine each model so that it fulfils the restraints dictated by the different experimental information. Furthermore, IMP has been implemented so that external users can also develop new restraints, optimization algorithms and analysis methods.