{
"title": "Model Fit Across Dopamine Computational Frameworks",
"metric": "Model fit (R² or variance explained)",
"papers": [
{
"doi": "10.1016/j.biopsych.2024.01.025",
"value": 0.69,
"condition": "Mouse, frontal cortex NE during threat learning",
"model_type": "Uncertainty temporal difference (NE threat prediction)",
"value_source_sentence": "We fit cue-evoked NE to the Rescorla-Wagner model and found that NE release fit this model both during 20 trials of tone/shock pairing (adjusted R 2 = 0.6869)"
},
{
"doi": "10.1016/j.biopsych.2024.01.025",
"value": "Rescorla-Wagner model fit: adjusted R² = 0.6869 (acquisition), adjusted R² = 0.6911 (extinction); behavioral freezing poorly predicted: adjusted R² = 0.0124 (acquisition), R² = 0.0241 (extinction)",
"condition": "mouse; optogenetics; GRABNE fluorescent sensor",
"model_type": "11.2",
"value_source_sentence": "We fit cue-evoked NE to the Rescorla-Wagner model and found that NE release fit this model both during 20 trials of tone/shock pairing (adjusted R 2 = 0.6869) and during 10 extinction trials (adjusted R 2 = 0.6911). In contrast, cue NE did not fully explain behavioral freezing to the same extent (ac"
},
{
"doi": "10.1038/s41467-024-51729-4",
"value": "Computational modeling reveals that these effects cannot be explained by increased decision noise but can be explained by value-independent risky bias and perseveration parameters, decision biases pre",
"condition": "rat; human; computational model",
"model_type": "11.3",
"value_source_sentence": "Computational modeling reveals that these effects cannot be explained by increased decision noise but can be explained by value-independent risky bias and perseveration parameters, decision biases previously linked to dopamine"
},
{
"doi": "10.3390/biomedicines13071783",
"value": "Machine learning models achieve a classification accuracy of 70-88% and may support the tracking of early treatment responses",
"condition": "rat; computational model; human fMRI",
"model_type": "11.5",
"value_source_sentence": "Machine learning models achieve a classification accuracy of 70-88% and may support the tracking of early treatment responses"
},
{
"doi": "10.1016/j.neuroimage.2026.121697",
"value": "<h4>Results</h4>The MCRNN achieved robust predictive performance (accuracy = 0",
"condition": "rat; human; computational model",
"model_type": "11.6",
"value_source_sentence": "<h4>Results</h4>The MCRNN achieved robust predictive performance (accuracy = 0"
},
{
"doi": "10.1371/journal.pcbi.1013226",
"value": "These same models also quantitatively matched mesolimbic dLight measurements better than non-Bayesian alternatives",
"condition": "mouse; rat; computational model",
"model_type": "11.4",
"value_source_sentence": "These same models also quantitatively matched mesolimbic dLight measurements better than non-Bayesian alternatives"
}
],
"n_analyzed": "Varies by study",
"description": "Comparison of model fit metrics (R², variance explained) across different computational accounts of dopamine: RL prediction error models vs. active inference precision models vs. Bayesian models",
"n_definition": "Varies: trials in behavioral tasks, neurons in recordings",
"scope_region": "Multiple brain regions (VTA, striatum, frontal cortex)",
"comparison_id": "fig_sec11_model_fit_comparison",
"taxonomic_level": "Not applicable (computational neuroscience)",
"scope_population": "Neural signals in model organisms and human behavior",
"homogeneity_check": "Low homogeneity: studies use different species (mouse, rat, human), different measures (neural firing, BOLD fMRI, neurotransmitter release), different model architectures. Direct comparison of R² values across studies is misleading without normalizing for task complexity and data type."
}