A simulated annealing algorithm for randomizing weighted networks.

Area of Science:

• Connectomics
• Network Science
• Computational Neuroscience

Background:

• Network null models are crucial for evaluating network features in scientific discovery.
• Current randomization methods in connectomics often preserve only binary node degrees, neglecting important weighted information.
• Advancements in imaging provide rich, biologically meaningful edge weights, necessitating improved analysis techniques.

Purpose of the Study:

• To introduce a novel simulated annealing procedure for generating randomized networks that preserve weighted degree (strength) sequences.
• To demonstrate the superiority and generalizability of this new method across various network types.

Main Methods:

• A simulated annealing algorithm was developed to generate randomized networks while preserving weighted degree sequences.
• Morphospace representation was employed to assess the algorithm's sampling behavior and ensemble variability.
• The proposed method was tested on diverse real-world network formats, including directed and signed networks.

Main Results:

• The simulated annealing procedure effectively preserves weighted degree sequences, outperforming existing rewiring algorithms.
• The method demonstrated generalizability across different network formats and real-world datasets.
• Accurate strength preservation led to distinct inferences regarding brain network organization compared to traditional methods.

Conclusions:

• The proposed simulated annealing method offers a powerful and simple approach for analyzing complex, next-generation connectomics datasets.
• This technique enhances the reliability of network feature evaluation by accurately accounting for connection weights.
• The findings underscore the importance of strength preservation for accurate inferences in brain network organization studies.