Comparative Study of a Variant Neural Relational Inference Deep Learning Model and Dynamical Network Analysis for...

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Protein allostery is crucial for biological regulation but complex to study.
• Dynamical network analysis and neural relational inference (NRI) are key computational methods.
• Understanding allosteric pathways is vital for deciphering protein function.

Purpose of the Study:

• To develop and evaluate a modified NRI model integrating transformer's multi-head self-attention for protein allostery.
• To compare the variant NRI model against dynamical network analysis and the initial NRI model.
• To investigate allosteric pathways in p53-DNA interactions, particularly comparing wild-type (WT) and mutant (MT) forms.

Main Methods:

• Modification of the NRI model by incorporating the multi-head self-attention module from transformers.
• Comparative analysis of the variant NRI model, dynamical network analysis, and the initial NRI model.
• Application to study allosteric interactions between p53 and DNA.

Main Results:

• The variant NRI model excels at predicting long-range allosteric pathways compared to dynamical network analysis.
• The modified NRI model demonstrates superior accuracy and comprehensiveness over the initial NRI model.
• Divergent allosteric pathways were observed between WT and MT p53, correlating with distinct DNA binding behaviors.
• Allosteric pathways propagate from mutation sites to DNA, with significantly longer pathways in MT p53.

Conclusions:

• The modified NRI model provides enhanced insights into protein allosteric pathways.
• Mutation sites in p53 impair long-range allosteric communication, affecting signal transmission.
• The findings offer a deeper understanding of protein allostery through advanced computational methods.