Neuron Platonic Intrinsic Representation From Dynamics Using Contrastive Learning

TL;DR

This work tackles extracting intrinsic neuron representations from multi-segment dynamics under varying peripheral conditions by positing a time-invariant, Platonic-like representation. It introduces NeurPIR, a contrastive learning framework based on VICReg that integrates peripheral information via CEBRA and uses segment-length augmentation to create robust positive pairs. The method demonstrates that learned embeddings encode neuron type and molecular information, and generalize to unseen animals, across both synthetic and real datasets. This approach enables robust, cross-modal analyses of neuronal systems and provides a scalable route for reinterpreting large neural datasets.

Abstract

The Platonic Representation Hypothesis suggests a universal, modality-independent reality representation behind different data modalities. Inspired by this, we view each neuron as a system and detect its multi-segment activity data under various peripheral conditions. We assume there's a time-invariant representation for the same neuron, reflecting its intrinsic properties like molecular profiles, location, and morphology. The goal of obtaining these intrinsic neuronal representations has two criteria: (I) segments from the same neuron should have more similar representations than those from different neurons; (II) the representations must generalize well to out-of-domain data. To meet these, we propose the NeurPIR (Neuron Platonic Intrinsic Representation) framework. It uses contrastive learning, with segments from the same neuron as positive pairs and those from different neurons as negative pairs. In implementation, we use VICReg, which focuses on positive pairs and separates dissimilar samples via regularization. We tested our method on Izhikevich model-simulated neuronal population dynamics data. The results accurately identified neuron types based on preset hyperparameters. We also applied it to two real-world neuron dynamics datasets with neuron type annotations from spatial transcriptomics and neuron locations. Our model's learned representations accurately predicted neuron types and locations and were robust on out-of-domain data (from unseen animals). This shows the potential of our approach for understanding neuronal systems and future neuroscience research.

Figures (3)

• Figure 1: Optimization objective for obtaining intrinsic representations of neurons as follows: clips(segments) from the same neuron should have a higher average similarity than clips from different neurons. Figure 1 illustrates how this objective can be achieved using a contrastive learning approach, where different recording segments from the same neuron are treated as positive pairs and segments from different neurons are treated as negative pairs. Each neuron is considered separately, and the Stimulus, Behavior and Session treated as surrounding information(SI), which acts as auxiliary variables. The surrounding information for each neuron is processed and encoded using CEBRA.
• Figure 2: a: the hyperparameters of five firing modes of neurons when simulating data; b: the neuron firing: the index of regular spiking (RS) is 0-400, the index of intrinsically bursting (IB) is 400-800, the index of chattering (CH) is 800-1200, the index of fast spiking (FS) is 1200-1600, the index of low-threshold spiking (LTS) is 1600-2000; c: the results of the visualization of neuronal activity data using UMAP.
• Figure 3: a: Steinmetz dataset contains 39 subdataset from 10 mice. Each subdataset records 400-700 neurons each from across the mouse brain during a visual behavior task. Each neuron with its brain area label. b: show of part of subdataset3. c: show of part of subdataset26.