Direct observation of the neural computations underlying a single decision

Natalie A Steinemann, Gabriel M Stine, Eric M. Trautmann, Ariel Zylberberg, Daniel M. Wolpert, Michael N. Shadlen

Abstract

Neurobiological investigations of perceptual decision-making have furnished the first glimpse of a flexible cognitive process at the level of single neurons. Neurons in the parietal and prefrontal cortex are thought to represent the accumulation of noisy evidence, acquired over time, leading to a decision. Neural recordings averaged over many decisions have provided support for the deterministic rise in activity to a termination bound. Critically, it is the unobserved stochastic component that is thought to confer variability in both choice and decision time. Here, we elucidate this drift-diffusion-like signal on individual decisions by recording simultaneously from hundreds of neurons in the lateral intraparietal cortex (LIP). We show that a single scalar quantity derived from the weighted sum of the population activity represents a combination of deterministic drift and stochastic diffusion. Moreover, we provide direct support for the hypothesis that this drift-diffusion signal is the quantity responsible for the variability in choice and reaction times. The population-derived signals rely on a small subset of neurons with response fields that overlap the choice targets. These neurons represent the integral of noisy evidence from direction-selective neurons within LIP itself. This parsimonious architecture would escape detection by state-space analyses, absent a clear hypothesis.

Introduction

For decades, neuroscientists have hypothesized that the brain makes decisions by accumulating noisy evidence until a threshold is reached—a process known as "drift-diffusion." While the Drift-Diffusion Model (DDM) is the gold standard in perceptual decision-making, its neural basis has largely been inferred from data averaged across many trials. This averaging suppresses the stochastic "diffusion" component, which is thought to be the primary source of variability in choice and timing. This study leverages high-density Neuropixels recordings in the lateral intraparietal cortex (LIP) to provide the first direct observation of these stochastic computations during individual decisions.

Experimental Paradigm and Theoretical Framework

The researchers trained rhesus macaques to perform a random dot motion discrimination task. In this task, monkeys must determine the net direction of moving dots and report their choice via a saccadic eye movement to one of two targets.

Figure 1: (a) Experimental task sequence showing fixation, target presentation, motion stimulus, and the final saccade. (c) Schematic of the drift-diffusion model where sensory evidence is integrated over time until it reaches a decision bound.

The DDM posits that the brain integrates "momentary evidence" (drift) plus "noise" (diffusion) over time. The challenge has been to resolve this "decision variable" (DV) in real-time during a single decision, rather than relying on trial-averaged ramps that might mask discrete jumps or alternative dynamics.

Methods

Using Neuropixels 1.0-NHP45 probes, the authors recorded simultaneously from up to 203 neurons in area LIP. This high recording density allowed for population-level analysis. They identified functional groups of neurons: $M_{in}$ neurons, which represent momentary sensory evidence, and $T_{in}$ neurons, which represent the integral of that evidence. To extract the decision signal, they employed various weighting schemes, including targeted regression, PCA, and simple functional grouping of $T_{in}$ neurons.

Behavioral Performance

The monkeys' behavior followed the classic patterns predicted by the DDM. As motion strength (coherence) increased, the monkeys became more accurate and faster, reflecting the more efficient accumulation of evidence.

Figure 2: Behavioral performance of two monkeys (M and J). Top: Reaction time as a function of motion strength. Bottom: Proportion of leftward choices as a function of motion strength.

Neural Dynamics of Decision Making

The study's key breakthrough is the direct observation of the decision variable on single trials. Several key insights emerged from the population analysis:

  • One-Dimensional Decision Signal: Despite the high dimensionality of the neural population, the decision-relevant signal is effectively a 1D scalar. Decoders trained at one time point performed consistently across others, suggesting a stable "coding direction."
  • Direct Observation of Diffusion: Single-trial traces of the population signal meandered and spread over time at a rate consistent with mathematical diffusion (variance proportional to time).
  • Mediation Analysis: Using mediation analysis, the authors showed that the signal at the end of a decision epoch accounted for the behavioral influence of all earlier samples. This provides "smoking gun" evidence that LIP neurons are indeed integrating evidence over time.
  • Functional Architecture: A small subset (~15%) of neurons with response fields overlapping the choice targets ($T_{in}$ neurons) carried nearly all the decision-relevant information. This suggests the "math" of the decision is computed through a parsimonious local circuit within LIP.

Conclusion

This research confirms that area LIP population activity directly encodes the decision variable that dictates both choice and timing. By observing the stochastic component of the neural signal on individual trials, the authors have validated the core tenets of the drift-diffusion model. While the study is correlational, it provides a remarkably clear view of the neural computations underlying a single cognitive act. Future work may explore whether these principles of local integration apply to more abstract, non-motor decisions.

Summary updated

Abstract

Neurobiological investigations of perceptual decision-making have furnished the first glimpse of a flexible cognitive process at the level of single neurons. Neurons in the parietal and prefrontal cortex are thought to represent the accumulation of noisy evidence, acquired over time, leading to a decision. Neural recordings averaged over many decisions have provided support for the deterministic rise in activity to a termination bound. Critically, it is the unobserved stochastic component that is thought to confer variability in both choice and decision time. Here, we elucidate this drift-diffusion signal on individual decisions. We recorded simultaneously from hundreds of neurons in the lateral intraparietal cortex of monkeys while they made decisions about the direction of random dot motion. We show that a single scalar quantity, derived from the weighted sum of the population activity, represents a combination of deterministic drift and stochastic diffusion. Moreover, we provide direct support for the hypothesis that this drift-diffusion signal approximates the quantity responsible for the variability in choice and reaction times. The population-derived signals rely on a small subset of neurons with response fields that overlap the choice targets. These neurons represent the integral of noisy evidence. Another subset of direction-selective neurons with response fields that overlap the motion stimulus appear to represent the integrand. This parsimonious architecture would escape detection by state-space analyses, absent a clear hypothesis.
BibTeX
@inproceedings{steinemann2024direct,
  author = {Natalie A Steinemann and Gabriel M Stine and Eric M. Trautmann and Ariel Zylberberg and Daniel M. Wolpert and Michael N. Shadlen},
  title = {{Direct observation of the neural computations underlying a single decision}},
  booktitle = {eLife},
  year = {2024},
  doi = {10.7554/elife.90859.3},
}
Figure