Neurotransmitter Classification
Image classification of synapses and neurons by neurotransmitter type (acetylcholine, GABA, glutamate) from electron micrographs, using a CNN trained in PyTorch.
Python PyTorch Deep Learning Neuroscience HPC
Overview
A deep learning pipeline to classify presynaptic terminals by neurotransmitter identity directly from EM image patches. Neuron-level predictions are derived by majority vote over a neuron’s constituent synapse predictions. The project explored both binary (excitatory vs. inhibitory) and three-class settings, with and without sensory neuron data.
Key Features
- VGG-based classifier in PyTorch with YAML-driven config and SLURM submission script
- Neuron consistency analysis on unlabelled data — >90% internal consistency across paired neurons in all setups
- Sensory neuron ablation study revealing their unexpected value for glutamate class separation