Current directions in combining simulation-based macromolecular modeling approaches with deep learning.

Area of Science:

• Computational biology
• Biophysics
• Drug discovery

Background:

• Structure-guided drug discovery relies on accurate computational macromolecular modeling.
• Current simulation methods have limitations in predicting folds, discovering function, and designing drugs.
• Deep neural networks (DNNs) have recently shown great promise in improving protein structure prediction accuracy.

Purpose of the Study:

• To introduce deep neural network architecture to biologists.
• To review deep learning's successes in protein structure prediction.
• To discuss emerging DNN-based methods for structure-function analysis and macromolecular design.

Main Methods:

• Review of recent literature on deep learning in macromolecular modeling.
• Focus on the integration of simulation-based and neural network-based approaches.
• Discussion of software architecture and code sharing for versatility.

Main Results:

• Deep learning has substantially enhanced accuracy in predicting protein structure from sequence.
• Emerging DNN-based approaches show potential for structure-function analysis and drug design.
• The synergy between simulation and deep learning is crucial for advancing the field.

Conclusions:

• Interconnection between simulation and DNNs is vital for macromolecular modeling.
• Modular software and open-source code are essential for versatile tool integration.
• Future drug discovery may leverage high-accuracy predictions from combined simulation and deep learning algorithms.