JuliaCon 2025

This work uses a novel computational approach, Neural Ordinary Differential Equations (NeuralODEs) with prospective configuration, to investigate the molecular mechanisms behind memory formation. For this purpose, this work concentrates on the modeling of the Synaptic Tagging and Capture hypothesis, one of the keystones that helps to understand how synaptic consolidation occurs during memory formation.
The STC hypothesis represents a fundamental mechanism in synaptic plasticity whereby synapses tag themselves by means of molecular tags, enabling LTP and subsequent memory consolidation. Indeed, traditional computational models of such biological processes have relied on backpropagation-based optimization algorithms. However, backpropagation's bidirectional gradient access methodology is in conflict with the known unidirectional nature of biological neural processes, which raises questions regarding its biological plausibility.
To overcome this limitation, prospective configuration utilizes a unidirectional optimization algorithm that more closely resembles biological neural processes. The combination of prospective configuration with NeuralODEs represents a new approach in computational neuroscience, as few studies have applied either technique to biological processes, particularly within the neuroscience domain.
The experimental framework is implemented in Julia because of its great scientific computing features and its efficiency in solving differential equations. NeuralODEs' implementation in the simulation of the STC hypothesis was particularly effective, resulting in a very low loss value of 0.005. A small loss would indicate that both temporal dynamics and molecular interactions could be well reflected by the model.
Systematic hyperparameter optimization identified a number of critical factors strongly influencing model performance. Network architecture was one of the crucial components, where certain configurations showed better performance in modeling the complex interactions involved in synaptic tagging. The selection of activation functions also turned out to be instrumental in improving the accuracy of the model with reduced loss values. These optimizations were necessary to arrive at a model that effectively captured the nuanced dynamics of synaptic plasticity and memory formation.
The findings contribute to a major breakthrough in both computational neuroscience and molecular biology. This research develops insights into the ways synaptic plasticity contributes to memory formation and strengthening of neural pathways by being able to successfully model the STC hypothesis through biologically plausible computational methods. The accuracy in the modeling of molecular interactions using this model makes it a good candidate for research in neurodegenerative conditions and memory-related disorders.
Success in this area will make NeuralODEs a very powerful tool for modeling such multifaceted biological processes. Their strength in representing continuous-time dynamics and molecular interaction with high precision will open up new avenues for research both in computational biology and in neuroscience. This approach bridges the gap between computational models and biological reality, allowing for a more faithful representation of neural processes than traditional artificial neural networks.
Besides, the methodology followed in the study provides a way in which future research into computational neuroscience could be conducted, especially in devising more biologically accurate models of neural processes. This is a promising direction for constructing computational models that reflect the underlying biology more closely, while keeping the computational efficiency: prospective configuration combined with NeuralODEs.
These studies have implications beyond the realm of theoretical understanding, pointing toward a possible development of therapeutic strategies in neurodegenerative diseases and disorders of memory. This work contributes to a more accurate computational model of synaptic plasticity and memory formation necessary for the foundation toward developing targeted interventions in conditions affecting memory and neural function.