Improving contact prediction along three dimensions.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Multiple sequence alignments (MSAs) of homologous proteins contain evolutionary information crucial for inferring 3D structures.
• Current methods typically involve sequence alignment, model selection, and parameter inference for contact prediction.

Purpose of the Study:

• To investigate the importance of each step in the contact prediction pipeline.
• To improve protein structure prediction by enhancing the predictive models used.

Main Methods:

• Developed and tested extensions to traditional pair-wise Potts models.
• Addressed challenges posed by gaps in MSAs.
• Evaluated performance on a large dataset of protein sequences.

Main Results:

• Demonstrated that all three dimensions of the prediction pipeline are critical for success.
• Showed significant improvements by advancing beyond pair-wise Potts models.
• Achieved combined improvements comparable to state-of-the-art methods.

Conclusions:

• Optimizing predictive models, particularly addressing MSA gap features, is key to advancing protein contact prediction.
• The proposed enhancements offer a substantial leap in inferring protein 3D structures from sequence data.