Research and Publications

Network Dynamics During Seizures.

While numerous novel treatments for epilepsy have been established over the past decades, the number of pharmacoresistant epilepsy patients has remained the same. This unresponsiveness to treatment suggests a fundamental lack of understanding of the physiological mechanisms related to intractable epilepsy. My Ph.D. thesis with Dr. Wim van Drongelen and postdoc with Dr. Catherine Schevon investigated the network dynamics and neural interactions across spatial scales in these patients. Using signal processing from human microelectrode array recordings from epilepsy patients undergoing intracranial monitoring, intracellular and extracellular recordings from acute brain slices, and computational modeling, I found evidence that neuronal saturation may explain failure of inhibition during seizures. Furthermore, I showed that neuronal saturation plays a strong role in the generation of high frequency activity in small networks. High frequency activity has been suggested as an electrophysiological marker to localize seizure foci on clinical recordings (e.g. EEG, ECoG), and my work showed that a majority of the high frequency activity observed at the clinical scale could be accounted for by volume conduction across spatial scales and harmonics caused by these small networks. Lastly, I found that localized, pathological neuronal firing during seizures has strong effects across the cortex and that the interactions between this localized activity and macroscopic cortical networks can describe the sustained oscillatory activity observed during seizures. This work implicated a dual role for inhibition during seizures, whereby brain regions that showed localized pathological firing were associated with failed inhibition, but macroscopic cortical networks required intact inhibition to produce oscillations. This novel scale- dependent view on inhibition during seizures has altered the belief that seizures require a failure of inhibition.

Inference with Asymmetric Evidence.

During my time working with Drs. Zachary Kilpatrick, Krešimir Josić, and Joshua Gold, I have investigated how observers learn to account for environmental structure and what sort of biases can occur when an observer is given evidence that should be weighted asymmetrically (where one type of evidence may be more/ less important to a decision). Using mathematical modeling approaches, I have developed Bayesian inference and heuristic models that can perform asymmetric evidence task. Notably, these models show significant choice asymmetries when evidence is asymmetric, even for the ideal Bayesian observer. I compared these results to human subjects recruited on the crowdsourcing platform Amazon MTurk who performed an asymmetric-evidence discrimination task and found that human subjects applied a variety of suboptimal strategies. Markedly, subjects who applied a suboptimal Bayesian approach increased their bias, responding with an enhanced choice asymmetry, but subjects who applied heuristics showed an increase in variance. Applying information theory metrics, I measured the complexity of subjects’ strategies and the proposed models and found that suboptimal models were typically less complex and less effective in their use of information. These results imply that the traditional bias-variance tradeoff inverts in the case of suboptimal performance for a given complexity level.

Learning Features of our Environment via Working Memory

Feature of our environment are often distributed heterogeneously, with particular values occurring more often (e.g., cardinal directions or certain colors). We can learn these environmental distributions in ways that help us to predict what features to expect in the future. Correspondingly, we show biases in favor of these common stimuli. We hypothesized that these biases are learned by the brain via experience, with changes to synaptic connectivity as we are exposed to stimuli over time, that reflect our beliefs about the probability we will observe a particular environmental feature. We built neural network models and simplified particle models to create theories of how the brain could implement learning and show human biases. We then compared learning and static models to human behavior and found that humans were better matched to learning models, suggesting that their biases could be the result of experience.

Future Directions

My long-term career goal is to apply a multimodal approach of theory, behavior, and advanced data analytical techniques on collaboratively collected human electrophysiology to understanding the neural and cognitive underpinnings that guide decision-making and related processes. My long term research goals are: 1) identifying cognitive strategies associated with flexible decision-making, 2) defining neural mechanisms that underlie latent environmental feature representation and integration with decision-making strategies, and 3) the impacts of pathological disruptions on these cognitive processes.

Additional Articles

Eissa TL. Mathematical Models of Decision-Making. Oxford Encyclopedia of Research (in review).

Cihak H, Eissa TL, Kilpatrick ZP. Distinct excitatory and inhibitory bump wandering in a stochastic neural field. SIAM Dyn. Syst. (2022).

Gill BJA, Wu X, Khan FA, Sosunov AA, Liou JY, Dovas A, Eissa TL, Banu MA, Bateman LM, McKhann GM II, Canoll PD, Schevon CA. Ex vivo Multi-Electrode Analysis Reveals Spatiotemporal Dynamics of Ictal Behavior at the Infiltrated Margin of Glioma. Neurobio. Dis. (2020).

Kilpatrick ZP, Holmes WR, Eissa TL, Josić K. Optimal Models of Decision Making in Dynamic Environments. Curr. Op. Neurobio. (2019).

Schevon CA, Tobochnik S, Eissa TL, Merricks EM, Gill BJA, Parrish RR, Bateman LM, McKhann Jr GM, Emerson RG, Trevelyan AJ. Multiscale Recordings Reveal the Dynamic Spatial Structure of Human Seizures. Neurobio. Dis. (2019).

Tryba AK, Merricks EM, Lee S, Pham T, Cho SJ, Nordli DR, Eissa TL, Schevon CA, van Drongelen W. The role of paroxysmal depolarization in focal seizure activity. J. Neurophys. (2019).

Eissa TL and Schevon CA. The Role of Computational Modeling in Seizure Localization. Brain (2017).

Canavan SV, Eissa TL, Schevon CA, McKhann GM Jr, Goodman RR, Emerson RG, van Drongelen W. Epileptogenic networks: applying network analysis techniques to human seizure activity. In: Validating Neuro-Computational Models of Neurological and Psychiatric Disorders (Basabdatta B, Fahmida C, ed): Springer. (2015)

Piochon C, Kloth AD, Grasselli G, Titley H, Nakayama H, Hashimoto N, Wan V, Simmons D, Eissa TL, Miyazaki T, Watanabe M, Takumi T, Kano M, Wang SH, Hansel C. Deficits in Cerebellar Plasticity and Motor Learning in a Copy Number Variation Mouse Model of Autism Spectrum Disorder. Nature Comm. (2014).

ElBasiouny SM, Quinlan KA, Eissa TL, Heckman CJ . Electrophysiological abnormalities in SOD1 transgenic models in Amyotrophic Lateral Sclerosis: the commons and differences. In: Amyotrophic Lateral Sclerosis (Maurer MH, ed): InTech (2011)