Integrating physics in deep learning algorithms: a force field as a PyTorch module.

Area of Science:

• Structural Biology
• Computational Chemistry
• Machine Learning

Background:

• Deep learning in structural biology is hindered by limited data, preventing convergence to accurate models.
• Current force fields are incompatible with deep learning due to implementation barriers.

Purpose of the Study:

• To introduce MadraX, a novel force field designed for seamless integration with deep learning.
• To enable end-to-end learning of physical principles in structural biology.

Main Methods:

• Implementation of MadraX as a differentiable PyTorch module.
• Integration of MadraX with deep learning frameworks for structural biology applications.

Main Results:

• MadraX facilitates end-to-end interaction between force fields and deep learning algorithms.
• Overcomes limitations of data scarcity in deep learning for structural biology.

Conclusions:

• MadraX provides a powerful tool for advancing deep learning applications in structural biology.
• The differentiable nature of MadraX enables more robust and accurate modeling.