De novo inference of protein function from coarse-grained dynamics

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Science

Background:

• Protein molecular function inference is crucial for understanding cellular processes but is challenging due to the vast number of newly discovered proteins.
• Current computational methods face limitations due to the lack of evolutionary information and high-throughput experimental techniques.

Purpose of the Study:

• To develop a de novo computational approach for annotating protein molecular function using structural dynamics matching.
• To address the limitations of existing methods by leveraging protein segment dynamics, even for proteins with low sequence identity.

Main Methods:

• A novel method matching structural dynamics of protein segments from dissimilar proteins (<10% sequence identity).
• Utilized 1 µs coarse-grained (CG) molecular dynamics trajectories and normalized root-mean-square-fluctuation graphs to identify mobile segments.
• Employed unweighted three-dimensional autocorrelation vectors for segment matching, using a custom-built forcefield (FF) validated against nuclear magnetic resonance data.

Main Results:

• Achieved an 87% true positive rate and 93.5% true negative rate on a dataset of 60 experimentally validated proteins.
• Demonstrated >99% true recall in a negative test against 315 unique fold/function proteins.
• Blind prediction on a novel protein showed consistency with additional retrieved evidence.

Conclusions:

• This work presents the first proof-of-principle for the generalized use of structural dynamics in inferring protein molecular function.
• The developed method, utilizing a custom CG FF, offers a powerful tool for protein function annotation, applicable to diverse proteins including moonlighting proteins and those with novel motifs.