Dissociable Microstructural Correlates of Learning Rate and Learning Noise in Gamified Reward-Based Decision-Making
Melina Vejlø, Niia Nikolova, Leah Banellis, Ashley Tyrer, Vasilisa Skvortsova, Tobias U. Hauser, Micah G. Allen
Public repository containing the data, source code, and figures for this project
Humans learn which actions yield the highest rewards through trial and error, gradually forming expectations about outcomes. Yet, people differ substantially in how quickly and precisely they learn. Such individual variability may partly be explained by differences in the brain’s microstructural organisation. In this large-scale study, 248 participants completed a gamified reward-learning task and underwent quantitative MRI to assess whole-brain microstructural indices of myelination (R1) and cortical iron (R2*). Using computational modelling, we quantified participants’ learning rates and learning noise, reflecting variability in how reward information is updated over time. Whole-brain voxel-based quantification analyses revealed that increased myelination in the cerebellum was associated with a higher learning rate, whereas learning noise was linked to increased myelination and iron concentration in the precentral gyrus. Together, these findings show that reward learning is not a unitary process but is instead shaped by distinct neurobiological pathways that support learning precision and noise. This work highlights how microstructural variation in sensorimotor and associative cortices contributes to stable versus variable reward learning behaviour across individuals.
├── README.md # Project overview
├── milkyway_vbq.Rproj # R project file file, for reproducibility of demographics data
├── docs # Directory containing demographics stats
│ ├── milky_way_demographics.qmd # Quarto document for demographics stats
├── vbq # Directory containing vbq analysis code
│ ├── run_analysis.m # wrapper script for running vbq analysis
│ └── mpm_vbq_runstats.m # vbq analysis function
├── data # Directory containing pre-processed behavioural RRST data
│ ├── milky_way_regressors.csv # Regressors file used for demographics (before map specific exclusions)
│ ├── MT_milky_way_regressors.csv # Regressors file used for vbq of MT map (after exclusions)
│ └── R1_milky_way_regressors.csv # Regressors file used for vbq of R1 map (after exclusions)
│ └── R2_milky_way_regressors.csv # Regressors file used for vbq of R2s map (after exclusions)If you use this code or data, please cite:
Vejloe, M., Nikolova, N., Banellis, L., Tyrer, A., Skvortsova, V., Hauser, T. U., & Allen, M. (2026). Dissociable Microstructural Correlates of Learning Rate and Learning Noise in Gamified Reward-Based Decision-Making. bioRxiv, 2026.02.24.707646. https://doi.org/10.64898/2026.02.24.707646