Computational Modeling Core_Frank

点击翻译按钮获取中文摘要

PROJECT SUMMARY (Computational Modeling Core, Core Leader: Frank, Brown University) The overarching goal of the Computational Modeling Core is to provide a common formal framework that can quantify dynamic decision processes in approach-avoidance conflict across species in Projects 1-4, including the impact of neural recordings and manipulations. We leverage hierarchical Bayesian parameter estimation of the drift diffusion model (HDDM), which captures not only choice proportions for varying reward, aversion, and conflict, but also the full response time distributions associated with these choices. HDDM facilitates reliable estimation of decision parameters and their modulation by trial-by-trial variance in neural signals, and supports Bayesian hypothesis testing for how these parameters may differ as a function of clinical status, brain state, and manipulations (e.g., nociceptin antagonism, acute/chronic stress, stimulation). We have shown how such “computational biomarkers” can provide enhanced sensitivity to discriminate between patient conditions and symptoms relative to traditional measures of behavior and brain activity. We will leverage neural recordings and stimulation from frontal cortex and basal ganglia across species to assess whether their variability is parametrically related to motivated evidence accumulation, and whether these signals are altered with neural manipulations and in clinical populations. Machine learning methods will quantify the degree to which such quantitative model fitting improves (1) classification of patient condition and brain state relative to the same methods applied to the raw behavioral and neural data or their summary statistics, and (2) our ability to map disease course, including suicidality and symptoms. Building on our extensive experience in neural networks and levels of computation involved in motivated learning and decision making across species, our computational framework will facilitate not only enhanced sensitivity to discriminate between clinical conditions, but will also identify hypotheses about the mechanisms involved, which will be tested via causal manipulations using the same quantitative framework. For example, our preliminary modeling studies indicate that variability in sub- populations within pregenual cingulate activity in non-human primates affects motivated evidence accumulation, and that in humans, the same parameter distinguishes MDD vs. healthy subjects and scales with symptoms. Moreover, these computational biomarkers are critical for predicting whether any individual is in one clinical state or another, whereas classification based on behavior and/or brain activity alone is at chance levels. The causal neural and psychological mechanisms of these effects will be further delineated and greatly expanded by utilizing the same quantitative framework with causal manipulations and more precise temporal recordings. Contribution to Overall Center Goals & Interactions with Other Center Components. As the Computational Modeling Core, our framework applies to approach-avoidance decision making across species and methods, and will be applied across all Projects. We will benefit from interactions amongst experts with complementary expertise in systems and cognitive neuroscience, psychiatry, computational modeling, and machine learning.