A computational neural model for mapping degenerate neural architectures

Zulqarnain Khan, Yiyu Wang, Eli Sennesh, Jennifer Dy, Sarah Ostadabbas, Jan-Willem van de Meent, Ben Hutchinson, Ajay B. SatputebioRxiv 2020 · DOI

Abstract

Degeneracy in biological systems refers to a many-to-one mapping between physical structures and their functional outcomes. Despite its ubiquity, traditional analytical tools for modeling degeneracy in neuroscience are extremely limited. In this study, we generated synthetic datasets to describe three situations of degeneracy in fMRI data to demonstrate the limitations of the current univariate approach. We describe a novel computational approach, Neural Topographic Factor Analysis (NTFA), designed to capture variations in neural activity across task conditions and participants. NTFA reveals whether and how experimental trials and participants cluster into task conditions and participant groups. We applied NTFA on simulated data, revealing the appropriate degeneracy assumption in all three situations and demonstrating NTFA’s utility in uncovering degeneracy.

Introduction

In cognitive neuroscience, the standard assumption is often a "one-to-one" mapping: a specific psychological task activates a specific set of brain regions. However, biological systems frequently exhibit degeneracy—a "many-to-one" mapping where multiple distinct structural configurations produce the same functional result. For instance, different individuals might employ different neural strategies to solve the same cognitive problem.

Traditional fMRI analysis methods, such as the General Linear Model (GLM), are inherently non-degenerate. They assume a single underlying activation pattern for a given task and treat deviations as random noise. Consequently, if degeneracy is present, the GLM "blurs" these distinct pathways into a single, misleading average map. To address this, we propose Neural Topographic Factor Analysis (NTFA), a generative model capable of discovering hidden, parallel neural architectures.

Methodology: Neural Topographic Factor Analysis (NTFA)

NTFA is a generative, primarily unsupervised computational model that treats fMRI data as a combination of three key components:

  1. Spatial Factors: The brain is modeled as a collection of "spatial factors," represented as radial basis functions (or "blobs") rather than individual voxels.
  2. Spatial Embeddings: These capture the precise location and shape variations of these factors across different participants.
  3. Activation Embeddings: These represent the magnitude of factor activation during specific trials for specific participants.

The Generative Approach: NTFA uses neural networks to map these low-dimensional embeddings to the actual spatial factors. This allows the model to capture non-linear relationships and ensures that the embeddings are "commensurable"—meaning researchers can directly compare the neural signatures of different participants in a shared visual space.

Experimental Validation

To validate NTFA against a known "ground truth," the researchers utilized synthetic fMRI datasets simulating three distinct scenarios:

  • Non-degeneracy: All trials and participants share one activation pattern.
  • Degeneracy by Condition: Different trials of the same task use different neural patterns (e.g., multiple ways the brain processes a single emotion).
  • Degeneracy by Participant and Condition: Patterns vary both by the specific trial and by the individual participant's strategy.

Key Results

The comparative analysis between the standard GLM and NTFA revealed significant differences in how they handle complex neural data:

  • GLM Limitations: In degenerate scenarios, the standard GLM produced an "amalgam" map—a blurry average that failed to represent any of the actual neural pathways used in the simulation.
  • NTFA Performance: NTFA successfully recovered the ground truth in all three scenarios. In the non-degenerate case, it identified two clean clusters (Baseline vs. Task). In the "Degeneracy by Condition" case, it revealed four clusters (one baseline and three task modes). Finally, in the "Participant/Condition" case, it correctly identified individual-specific neural strategies.

Discussion and Future Work

The most profound insight from this work is that what neuroscientists currently discard as "residual error" or "noise" might actually contain structured evidence of degeneracy. NTFA provides a discovery-oriented framework that does not require prior knowledge of trial conditions, instead "discovering" clusters in the embedding space.

While this study relies on simulated data, it lays the groundwork for applying NTFA to large-scale, real-world fMRI datasets. Future research will investigate whether previously "unreliable" findings in cognitive neuroscience are, in fact, evidence of diverse, degenerate neural strategies across different populations.

Summary updated

Abstract

Abstract Degeneracy in biological systems refers to a many-to-one mapping between physical structures and their functional (including psychological) outcomes. Despite the ubiquity of the phenomenon, traditional analytical tools for modeling degeneracy in neuroscience are extremely limited. In this study, we generated synthetic datasets to describe three situations of degeneracy in fMRI data to demonstrate the limitations of the current univariate approach. We describe a novel computational approach for the analysis referred to as neural topographic factor analysis (NTFA). NTFA is designed to capture variations in neural activity across task conditions and participants. The advantage of this discovery-oriented approach is to reveal whether and how experimental trials and participants cluster into task conditions and participant groups. We applied NTFA on simulated data, revealing the appropriate degeneracy assumption in all three situations and demonstrating NTFA’s utility in uncovering degeneracy. Lastly, we discussed the importance of testing degeneracy in fMRI and the implications of applying NTFA to do so.
BibTeX
@misc{khan2020computational,
  author = {Zulqarnain Khan and Yiyu Wang and Eli Sennesh and Jennifer Dy and Sarah Ostadabbas and Jan-Willem van de Meent and Ben Hutchinson and Ajay B. Satpute},
  title = {{A computational neural model for mapping degenerate neural architectures}},
  year = {2020},
  doi = {10.1101/2020.11.13.382192},
  howpublished = {bioRxiv},
}
Figure