Dissecting Neural Contributions to Decision-Making: The Contribution of Brain Regions and Neuron Subtypes in Choice Prediction
The prediction of an animal's choice from neural activity is crucial for understanding decision‐making and developing brain–computer interfaces. We here applied the Steinmetz et al. (2019) Neuropixels dataset of mice trained on a two‐alternative visual discrimination on a steering‐wheel task. We focused on five regions - spanning hippocampal, cingulate, and motor areas: dentate gyrus (DG), CA3, anterior cingulate area (ACA), primary motor cortex (MOp) and secondary motor cortex (MOs). We first trained time‐resolved, cross‐validated logistic‐regression classifiers within sliding 300 ms windows (50 ms steps) to decode both the direction (left vs. right) and correctness of each choice from population spiking. Accuracy for choice decoding rose systematically from the hippocampal subfields (median ≈ 0.55–0.58) to ACA (≈ 0.70) to motor areas (MOs/MOp ≈ 0.80–0.88), whereas trial outcome decoding was uniformly high (≈ 0.75–0.85) across regions. Onset‐latency analysis revealed that ACA and DG were the only ones with discrete, post‐stimulus choice accuracy increases above pre‐stimulus baseline + 2 × SEM (occurring at 245 ms for ACA and 295 ms for DG). Cross‐temporal generalization revealed that the hippocampus contained fleeting choice representations, whereas the motor cortex exhibited extended, time-invariant coding, as would be expected from downstream motor‐planning signals. Finally, contrasting high‐ versus low‐contrast trials as a proxy for task difficulty, we found that stimulus strength modulates the timing and reliability of choice signals. Specifically, higher contrast stimuli accelerates and stabilizes decision‐related activity along this ACA→motor axis. Together, these results describe a hierarchical, contrast‐sensitive flow of choice information from medial temporal through prefrontal to motor circuits during perceptual decisions.
Institutions
- Universidade Federal de Minas Gerais (BR)
- Tarbiat Modares University (IR)
- University of California, Los Angeles (US)
- Allen Institute (US)
- International Trachoma Initiative (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2025-04-14
- DOI
- https://doi.org/10.5281/zenodo.15126428
- Primary Topic
- Graph Theory and Algorithms
- Type
- article
- Field-Weighted Citation Impact
- 0.00