Summary: A new open-source software framework integrates dendritic properties and mechanisms into large-scale neural network models to improve biological realism and support advances in AI and neuroscience.
Source: FORTH
Researchers at FORTH-IMBB have developed a practical computational toolkit to include dendritic structure and function in large-scale neural simulations, enabling new investigations into how dendrites shape information processing in the brain.
The toolkit, released as an open-source Python package, makes it possible to incorporate essential dendritic properties into spiking neural network models without prohibitive computational cost. This advance has implications for basic neuroscience, the study of neurological disease, and the design of more biologically informed artificial intelligence systems.
Understanding how the brain computes remains one of the major scientific challenges of the 21st century. Progress in modeling the cellular and subcellular mechanisms that support cognition, learning and memory can inform treatments for neurological disorders and inspire novel AI architectures that mimic brain efficiency and adaptability.
In a recent publication in Nature Communications, the team led by Research Director Dr. Panayiota Poirazi at the Institute of Molecular Biology and Biotechnology (IMBB) of the Foundation for Research and Technology – Hellas (FORTH) introduces Dendrify, a framework designed to bring dendritic computations into large-scale spiking neural network models.
Dendrites are the tree-like extensions of neurons that receive and integrate synaptic inputs. Historically, their role in computation remained incompletely understood because experimental and computational tools were limited. Recent work, however, has revealed that dendrites host a range of nonlinear mechanisms capable of performing complex local computations and contributing to synaptic plasticity—the cellular basis of learning and memory.
Although the contribution of dendritic processes to the behavior of single neurons is increasingly well described, their effects at the circuit and network level remain largely unexplored. Dendritic structure and function change with aging and in neurodegenerative diseases such as Alzheimer’s, and some studies have linked dendritic complexity to cognitive measures. Therefore, tools that scale dendritic mechanisms to network-level models are needed to investigate their system-wide roles.
Dendrify was created to bridge the gap between two common but limited approaches: simple spiking neural networks that are efficient but neglect dendritic dynamics, and fully detailed morphologically realistic neuron models that are biologically accurate but computationally expensive for large networks. Dendrify automatically generates reduced compartmental neuron models that retain key dendritic and synaptic integration features while remaining computationally tractable.
By offering a balance between biological realism, flexibility, and performance, Dendrify enables researchers to explore how dendritic mechanisms influence network-level computations and cognitive functions. The framework is designed to be user-friendly so that even users without extensive modeling experience can build neuronal populations with dendritic compartments and examine their effects on activity, plasticity, and information processing.
Beyond basic research, these models can inform the development of neuromorphic hardware and neuro-inspired algorithms that leverage dendritic-like computation for more efficient and adaptive artificial intelligence systems. Several recent AI studies have drawn inspiration from dendritic processing to improve learning rules and network architectures, and Dendrify makes it easier to test such ideas at scale.
This effort was led by Michalis Pangalos, a Ph.D. candidate in the Department of Biology at the University of Crete, in collaboration with Dr. Spiros Chavlis, a postdoctoral researcher at IMBB, under the supervision of Dr. Poirazi. The work aims to make dendritic modeling accessible and widely usable in both neuroscience and AI research communities.
About this AI and neuroscience research news
Author: Press Office
Source: FORTH
Contact: Press Office – FORTH
Image: The image is in the public domain
Original Research: Open access.
Title: “Introducing the Dendrify framework for incorporating dendrites to spiking neural networks” by Michalis Pagkalos et al., Nature Communications
Abstract
Introducing the Dendrify framework for incorporating dendrites to spiking neural networks
Computational modeling is essential for understanding how subcellular neuronal features, such as dendritic mechanisms, influence circuit processing. Yet the network-level consequences of dendritic computations have been difficult to address because existing tools either ignore dendrites or are too computationally demanding for large-scale simulations.
Current spiking neural network models are often efficient but lack important dendritic properties, while morphologically detailed neuron simulations provide biological accuracy at the cost of prohibitive computational load. Dendrify addresses this trade-off by providing an open-source Python package based on the Brian 2 simulator that generates reduced compartmental neuron models using simple commands. These models capture biologically relevant dendritic and synaptic integration features while remaining efficient for network simulations.
By lowering the barrier to include dendritic features in spiking neural networks, Dendrify enables systematic studies of how dendrites contribute to emergent network behaviors and supports the design of neuromorphic systems and AI algorithms that benefit from dendrite-inspired computation.