Deep neural network processing of DEER data.

Area of Science:

• Biophysics
• Spectroscopy
• Computational Chemistry

Background:

• Model-free methods, including DEER, PELDOR, DQ-EPR, and RIDME, are standard for analyzing two-electron dipolar spectroscopy data.
• These established techniques typically rely on regularized fitting for data processing.

Purpose of the Study:

• To explore the application of artificial neural networks (ANNs) for processing DEER data.
• To evaluate the accuracy and reliability of ANNs compared to traditional methods.

Main Methods:

• Development of ANNs trained on extensive databases of simulated two-electron dipolar spectroscopy data.
• Implementation and testing of these ANNs on real experimental DEER data.
• Incorporation of the developed neural networks into established software packages (Spinach and DeerAnalysis).

Main Results:

• ANNs demonstrated unexpectedly high accuracy and reliability when processing real experimental DEER data.
• The neural networks effectively identified and rejected exchange interactions.
• The ANNs provided a quantitative measure of uncertainty in the resulting distance distributions.

Conclusions:

• Artificial neural networks present a powerful and reliable alternative for analyzing DEER data.
• ANNs offer advantages over traditional methods by handling complex interactions and providing uncertainty estimates.
• The integration of these ANNs into Spinach and DeerAnalysis enhances their analytical capabilities for EPR spectroscopy.