Enhancing evaluation of reward learning using computational modeling methods

研究经常依赖于行为任务作为理解病因过程，相关因素， 精神病理学的后果在奖励学习和决策的背景下，爱荷华州赌博 任务（IGT）经常使用。最终，研究通常依赖于任务的原始执行， 允许参与者在任务中充分指导探索和学习。这降低了研究人员的能力， 区别奖励和惩罚的学习。此外，尽管IGT涉及多个过程， 性能，任务经常被概括为单个性能度量（即，播放比例 有利/不利的刺激）。更现代的方法，包括计算建模， 提供了区分进程的手段（例如，奖励学习，惩罚学习，奖励 敏感性，持续倾向），这是感兴趣的实质性研究问题，改变 在精神病理学中起作用该领域的一个关键限制是， IGT的建模方法仅使用了任务的原始版本。更强的实验 范例可以增加估计参数的精度。此外，计算机建模研究 IGT，特别是在发育样本中，依赖于横截面设计，排除了检查 重测信度或纵向变化。这个项目将估计IGT的计算模型， 一个更新版本的任务，具有完整的实验控制的刺激呈现在所有试验。在 独立样本（研究1 n = 50名本科生;研究2 [R 01 MH 107495]后代和父母（n = 248），IGT在多个场合施用（研究1两次，大约间隔4周;研究2向上 五次，后代和父母间隔约9个月）。使用研究1的数据，我们将 开发奖励学习计算模型来描述IGT更新版本中的任务行为。 使用研究2的数据，我们将检验模型对独立样本的可推广性。我们将 使用计算建模方法检查性能的重测信度，但扩展 重复评估次数。我们还将检查模型参数对指数的有效性， 奖励功能测量（包括后代和父母的自我报告;后代的脑功能; 抑郁症家族史）。最后，我们还将调整计算模型，以包括 发展，以模拟后代样本的变化。除了对《联合国宪章》的实质性贡献外， 通过这些任务和建模调整来评估奖励学习和决策，我们还将使用 我们的工作是通过更新开源软件（hBayesDM）来帮助传播这些模型 包），以包括我们的模型供其他研究人员使用。

Studies of frequently rely on behavioral tasks as means of understanding etiological processes, correlates, and consequences of psychopathology. In the context of reward learning and decision making, the Iowa Gambling Task (IGT) is frequently used. Ultimately, studies routinely rely on the original implementation of the task that permits participants to fully direct the exploration and learning in the task. This reduces researchers' ability to distinguish reward and punishment learning. Moreover, despite multiple processes being involved in IGT performance, the task is frequently summarized as a single performance metric (i.e., proportion of plays on advantageous/disadvantageous stimuli). More modern approaches, including computational modeling, provides a means to distinguish between processes (e.g., reward learning, punishment learning, reward sensitivity, perseveration tendency) that are of interest to substantive research questions about altered functioning in psychopathology. A critical limitation of the field is that studies that have used computational modeling approaches with the IGT have only used the original version of the task. A stronger experimental paradigm may increase the precision of the estimated parameters. Moreover, computational modeling studies of the IGT, particularly in developmental samples, rely on cross-sectional designs that precludes examining test-retest reliability or longitudinal change. This project will estimate computational modeling of the IGT using an updated version of the task that has full experimental control of the stimulus presentation across all trials. In independent samples (Study 1 n = 50 undergraduates; Study 2 [R01 MH107495] offspring and parents (n = 248), the IGT was administered on multiple occasions (Study 1 twice, approximately 4 weeks apart; Study 2 up to five occasions, approximately 9 months apart for offspring and parents). Using data from Study 1, we will develop reward learning computational models to describe task behavior in the updated version of the IGT. Using data from Study 2, we will examine the generalizability of the model to independent samples. We will examine test-retest reliability of performance using the computational modeling approaches, but expand the number of repeated assessments. We will also examine the validity of model parameters against indices of reward functioning measures (including self-reports in both offspring and parents; brain function in offspring; and family history of depression). Finally, we will also adapt the computational model to include indices of development to model changes in the offspring sample. In addition to the substantive contributions to the assessment of reward learning and decision making by these task and modeling adaptations, we will also use our work to aid in the dissemination of these models by updating open source software (the hBayesDM package) to include our models for use by other investigators.