Research & Projects

Located within the basal ganglia, the striatum is fundamental to processes like motor control and reward-based learning. Recent studies highlight an intricate relationship between dopamine, striatal astrocytes, in the striatum. We propose that, via gliotransmitter release or dopamine modulation, astrocytes might play a pivotal role in shaping striatal pathways, ultimately influencing decision-related behaviors.

Methods

Using optogenetic techniques, we stimulated dopamine neurons in the SNpc while concurrently monitoring astrocyte activity. We utilized two-color photometry imaging to simultaneously capture dopamine and astrocyte activities in the dorsolateral and dorsomedial striatum of free-moving mice during a maze-based decision-making task. DeepLabCut-driven behavioral recordings enabled precise behavior categorization.

Results

The data revealed a two-way relationship between dopamine and astrocytes, with surges in astrocyte calcium levels following dopamine activity peaks. Moreover, decreasing dopamine activity correlated with increased astrocyte responses. Our results demonstrate astrocytes' significant reactivity to optogenetically-induced dopamine release. Notably, we observed a decline in dopamine levels following astrocyte optogenetic activation. Furthermore, during the switch in behavioral states—from disengaged to engaged—we observed heightened astrocyte calcium activity. These calcium peaks in astrocytes correlated with the onset of active engagement, hinting at astrocytes' possible role in initiating engagement.

To understand basal ganglia functions, we need to understand the output functions of these forebrain systems. Much is known about the direct and indirect pathways that originate in the striatal matrix and end in the brainstem motor nuclei, but little is known about the circuits underlying non-motor functions of the basal ganglia. Previous work, however, has suggested that striosomes could be important for these. They are preferentially innervated by limbic-related regions of neocortex and by some limbic structures such as the bed nucleus of the stria terminals.

Methods

Using in vivo two-color photometry calcium imaging, we monitor the dynamics of S-D1 and S-D2 neurons, as well as dopamine release, in pairs of simultaneous recordings in the dorsolateral striatum (DLS) and the dorsomedial striatum (DMS), using transgenic mice and viral methods during a probabilistic T-maze decision-making task. 

I have developed a comprehensive computational analysis pipeline, using unsupervised clustering to dissect the behavior of the mice frame by frame, and have implemented reinforcement learning (RL) algorithms to calculate action values from past rewards. I have created a package to synchronize various trial events, considering differing intra-event intervals. The approach facilitates a detailed, nuanced examination of the influence of action value on neuronal activity and dopamine release, while effectively isolating these effects from those of movement.

Results

Only S-D1 neuronal activity, not S-D2, demonstrates pronounced peaks at both the beginning and end of the trials, which might have particular importance in the transition from goal-driven to a more nearly ‘packaged,’ chunked habitual status of the runs practiced. 

Suicide is among the leading causes of death in youth worldwide. Early identification of children with high risk for suicide is a key to effective screening and intervention strategies. Yet, little is known about the neural pathways to the clinical outcomes of youth suicide

Methods

Using fMRI and self-questionnaire data from the largest ongoing study of brain development and children’s mental health in the United States (Adolescent Brain Cognitive Development study), I developed a Machine Learning (ML) model that sought potential predictors for suicidal tendencies in children.

Results

The findings demonstrated that, in the ML model designed to predict childhood suicidality, the feature of highest importance was the activation level of the anterior cingulate cortex during the stop-signal task followed by established risk factors such as emotion regulation, impulsivity, and family environment in terms of feature importance.

[Link to: Code, Ahn et al., Studies in Computational Intelligence 2021 ]

In this retrospective study, 10,000 anteroposterior (AP) radiography of the knee from a single institution was used to create a medical data set that is more balanced and cheaper to create.

Methods

By training Generative Adversarial Networks (GANs) on unlabeled X-ray images and employing Principal Component Analysis (PCA) within the GAN's latent space, I distinguished the principal components that signified the size, orientation, and osteoarthritis progression in knee images. I concentrated on creating images that depict the advancement of arthritis by adjusting the latent vectors associated with the third principal component.

Results

Using a GAN, we were able to generate knee X-ray images that accurately reflected the characteristics of the arthritis progression stage, which neither human experts nor artificial intelligence could discern apart from the real images. In summary, our research opens up the potential to adopt a generative model to synthesize realistic anonymous images that can also solve data scarcity and class inequalities.

[Link to: Code, Ahn et al., Journal of Orthopaedic Research 2022 ]