关键词: 决策偏差, 腹侧前额叶皮层, 岛叶, 漂移扩散模型, 风险偏好

Abstract: Decision biases, defined as a systematic deviation from rational choice in risky contexts, are fundamental features of human cognition. Although phenomena like the framing effect are well-documented, yet the computational and neurobiological bases underlying these biases remain unclear. This study aimed to characterize multilevel framework of decision bias by integrating behavioral experiments, computational modeling, neuroimaging, and transcriptomic analyses. We conducted five independent studies involving a large combined sample size (total N = 1, 257) by using a classic risky choice paradigm (Cups Task). To ensure the robustness of findings, we controlled for potential effects of sex, reward magnitude, and decision pressure. In a subset of participants, functional magnetic resonance imaging (fMRI) was employed during the task to capture the brain regions responding to experienced risk and reward. The Drift Diffusion Model (DDM) was applied to estimate key computational parameters (drift rate and non-decision time) for each participant and across the gain and loss domains, reflecting the underlying decision dynamics. To bridge the neural and molecular levels, imaging–transcriptomic analyses were employed to spatially map gene expression patterns associated with task-related brain regions. A robust and replicable decision bias was found across all five studies: participants consistently exhibited greater risk-seeking in the gain domain compared to the loss domain, accompanied by faster responses. The DDM revealed significantly higher drift rates and shorter non-decision times in the gain domain, with the model-derived decision bias index strongly correlated with behavioral biases. Neuroimaging results further revealed greater ventromedial prefrontal cortex (vmPFC) activation during reward experiences and reduced superior parietal lobule (SPL) engagement during risk processing in the gain domain. The bilateral anterior insula (aINS) tracked both risk and reward experiences and predicted individual decision biases; notably, aINS activation differentially encoded domain-specific risk, and its activity was found to be fully mediated by the DDM-derived computational bias index in predicting the behavioral bias index. The amygdala showed no involvement in decision biases, indicating limited engagement of emotional reactivity. Finally, spatial transcriptomic analyses identified distinct gene expression signatures associated with decision-bias-related activation patterns, enriched for ion transport and synaptic processes, with cell-type specificity in excitatory/inhibitory neurons and astrocytes. Neurotransmitter mapping revealed associations with dopaminergic (D1), serotonergic (5HTT), and noradrenergic (NET) receptor distributions. In conclusion, these findings reveal convergent behavioral, computational, neural, and transcriptomic associations related to decision bias in risky decision-making, providing a multilevel characterization of the processes that may contribute to individual differences in risk preferences.

Key words: decision biases, vmPFC, insula, DDM, risk preference