Tracking the Fidelity of Internal Neural Representations with Error-In-Variables Regression.

Area of Science:

• Systems Neuroscience
• Computational Neuroscience
• Neuroscience

Background:

• Internal neural representations often deviate from external sensory and behavioral measurements.
• A lack of principled statistical frameworks hinders the quantification of these neural representation mismatches.

Purpose of the Study:

• To introduce a novel nonlinear error-in-variables regression framework.
• To provide a statistical tool for quantifying the fidelity of internal neural representations.

Main Methods:

• Developed a nonlinear error-in-variables regression framework.
• Employed flexible basis expansion and sampling-based inference.
• Jointly inferred neuron-specific tuning functions, latent trajectories, and representational fidelity parameter (κ).

Main Results:

• The model accurately recovered latent dynamics and tuning curves on synthetic data.
• Identified condition-dependent changes in representational fidelity and tuning profiles in mouse and rat brain recordings.
• Uncovered latent population manifolds obscured by conventional analyses.

Conclusions:

• The developed framework offers a powerful and computationally tractable method for analyzing neural representation fidelity.
• This approach advances systems neuroscience by enabling precise tracking of internal neural dynamics and their relation to behavior.