NPEX: Never give up protein exploration with deep reinforcement learning

Area of Science:

• Computational Biology
• Structural Biology
• Artificial Intelligence

Background:

• Determining unknown protein structures, including metastable states, is crucial for therapeutic agent design.
• Current computational methods like molecular dynamics and Markov chain Monte Carlo simulations are time-consuming and require prior structural information.

Purpose of the Study:

• To develop an innovative and efficient method for protein structure determination.
• To overcome the limitations of existing computational approaches in terms of speed and data requirements.

Main Methods:

• Developed the never give up protein exploration (NPEX) method utilizing deep reinforcement learning.
• Employed the soft actor-critic algorithm and an intrinsic reward system to introduce bias potential without prior knowledge.
• Applied NPEX to benchmark models: double well, triple well, alanine dipeptide, and tryptophan cage.

Main Results:

• NPEX demonstrated markedly greater sampling efficiency compared to Markov chain Monte Carlo simulations.
• The method effectively determined protein structures without necessitating prior domain knowledge.
• Achieved significantly enhanced computational efficiency in protein structure exploration.

Conclusions:

• The NPEX method offers a revolutionary approach to protein structure determination.
• Its enhanced computational efficiency and independence from prior knowledge will accelerate drug discovery and fundamental protein research.