Related Experiment Video
Updated: Jun 13, 2025
12:27
Large-scale Reconstructions and Independent, Unbiased Clustering Based on Morphological Metrics to Classify Neurons in Selective Populations
Published on: February 15, 2017
6.9K
Learning to cluster neuronal function
Nina S Nellen1, Polina Turishcheva1, Michaela Vystrčilová1
1Institute of Computer Science and Campus Institute Data Science, University Göttingen, Germany.
Arxiv
|June 12, 2025
Summary
This study introduces DECEMber, a novel method to improve cell type identification in the visual cortex using deep learning. DECEMber enhances neuron clustering in digital twins of the brain, revealing clearer functional organization.
Area of Science:
- Computational Neuroscience
- Machine Learning
- Neuroscience
Background:
- Deep neural networks (DNNs) show promise as digital twins of the visual cortex, enabling per-neuron embedding analysis.
- Current DNNs for mouse V1 lack clear per-neuron embedding clusters, questioning model limitations versus biological reality.
Purpose of the Study:
- To develop a method that enhances clustering of per-neuron embeddings from DNNs.
- To investigate the functional organization of the mouse V1 by improving cell type identification.
Main Methods:
- Introduced DECEMber (Deep Embedding Clustering via Expectation Maximization-based refinement), incorporating an auxiliary loss function for structured embeddings.
- Jointly optimized neuronal feature embeddings and clustering parameters using an EM-algorithm.
Main Results:
- DECEMber improved cluster consistency and stability compared to standard methods.
- The method maintained high predictive performance of the DNNs.
- DECEMber demonstrated generalization across species (mice, primates) and visual areas (retina, V1, V4).
Conclusions:
- DECEMber effectively enhances the structured organization of per-neuron embeddings in DNNs.
- The findings suggest that improved modeling can reveal clearer functional cell type organization in the visual cortex.
- The approach is robust and applicable to diverse neural data.
