Limits on the computational expressivity of non-equilibrium biophysical processes

Area of Science:

• Biophysics
• Computational Biology
• Biochemistry

Background:

• Biological decision-making often involves classification tasks using biophysical processes.
• Understanding computational limits of biological systems is crucial for designing physical computing.

Purpose of the Study:

• Investigate computational expressivity of biochemical networks for classification.
• Identify limitations and potential enhancements in biophysical computing.

Main Methods:

• Modeling biochemical networks as Markov jump processes.
• Training these networks for classification tasks.
• Analyzing decision boundary flexibility, sharpness, and classification capacity.

Main Results:

• Revealed unanticipated limitations in input-output functions of biochemical networks.
• Demonstrated that promiscuous binding can overcome these limitations.
• Identified network signatures like correlated spanning trees and creased energy landscapes.

Conclusions:

• Biophysical media impose specific constraints on computational expressivity for classification.
• Biochemical mechanisms can be leveraged to enhance the computational power of these systems.
• Findings inform the design of biological and synthetic physical computing systems.