Project 1: Neural Representation of Abstract Rules in the Hippocampus
Understanding how the hippocampus encodes abstract concepts such as rules is crucial for deciphering cognitive flexibility. In this project, we combine high-throughput behavioral tasks with calcium imaging of medial temporal lobe areas to investigate how rule-based memory guides complex task resolution. By applying Hidden Markov Models to segment behavioral data over time, we correlate neural activity patterns with distinct rule implementations, unveiling the neural dynamics underlying abstract rule representation.
Project 2: Disentangling Memory and Actions from Behavior Using POMDPs
To understand how animals balance memory and decision policies in dynamic environments, we model behavior in our 8-port maze task using Partially Observable Markov Decision Processes (POMDPs). By simulating agents under different environmental conditions with where rule volatility changes in real time, we aim to disentangle the contribution of memory expressed in the a priori internal believes, from policy adaptation during task resolution. Our approach integrates mice behavioral data fitting with computational modeling to reveal how animals update strategies in response to changing uncertainty.
Project 3: Large Language Models for Digital Twins in Drug Discovery
This project leverages large-scale data banks to develop large language models (LLMs) for creating digital twins in drug discovery and testing. By integrating vast biological and pharmacological datasets, our models aim to simulate drug interactions, predict therapeutic outcomes, and optimize experimental designs. This approach enhances precision medicine by enabling in silico testing, accelerating hypothesis generation, and reducing reliance on traditional trial-and-error methodologies.
Project 4: Exploring the Dimensionality of Spatial Representations in the Dorsal Hippocampus
Using electrophysiological recordings from freely moving mice, we investigate the high-dimensional structure of neuronal representations in the dorsal hippocampus. Building on our discovery that place-cell responses are modulated by heading direction relative to reference points, we aim to identify additional behavioral variables that shape neuronal activity. By analyzing how these representations evolve as animals learn to navigate novel environments, we seek to understand how preexisting network structures support spatial cognition and adaptive behavior