adult-neurogenesis-olfactory-representational-stability

name	adult-neurogenesis-olfactory-representational-stability
description	Adult-neurogenesis dual role methodology — spiking network model showing how continuous addition of new neurons supports both odor representational stability and flexibility in olfactory circuits.
tags	["neuroscience","neurogenesis","olfactory-system","representational-drift","spiking-network","computational-neuroscience","plasticity","neural-coding","brain-network"]
created	2026-05-27T00:00:00.000Z
source	DOI: 10.7554/eLife.107905 | PMID: 42112574

Adult-Neurogenesis Allows for Representational Stability and Flexibility in Early Olfactory System

Overview

This methodology from Chen & Padmanabhan (eLife, 2026) uses a detailed spiking network model of early olfactory circuits to reveal how adult neurogenesis (continuous addition of new neurons throughout life) simultaneously enables two seemingly opposing properties:

1. Representational stability: faithful odor encoding at the population level
2. Representational flexibility/drift: experience-dependent plasticity and learning

The model covers two major olfactory processing stages with distinct computational roles.

Core Model Architecture

Stage 1: Main Olfactory Bulb (MOB)

• Adult neurogenesis affects individual cell responses but preserves population-level representations
• New neurons (granule cells) provide inhibitory interneuron replacement
• Net effect: individual mitral/tufted cells shift, but population code remains robust
• Mechanism: lateral inhibition redistribution via new granule cells

Stage 2: Piriform Cortex (PCx)

• Both individual cell responses AND population dynamics undergo progressive change
• Representational drift: stimulus-evoked activity patterns gradually change
• Drift rate is experience-dependent — repeated odor exposure reduces drift
• Implements a form of temporal context coding

Key Findings

1. MOB preserves population code: Even as individual neurons are replaced/rewired, the high-dimensional population vector representation of each odor remains stable
2. PCx implements representational drift: The cortex continuously updates its odor representations — encoding not just what but when
3. Experience protects stability: Frequently encountered odors have more stable representations (reduced drift)
4. Dual functional role: Same neurogenesis process serves both stability (MOB) and flexibility (PCx) via different circuit mechanisms

Spiking Network Model

import numpy as np
from typing import List, Tuple

class OlfactoryBulbModel:
    """Simplified spiking network model of main olfactory bulb."""
    
    def __init__(self, n_glomeruli=200, n_mitral=200, n_granule=1000,
                 neurogenesis_rate=0.01):
        self.n_glom = n_glomeruli
        self.n_mitral = n_mitral
        self.n_granule = n_granule
        self.neurogenesis_rate = neurogenesis_rate  # Fraction replaced per day
        
        # Connectivity matrices
        self.W_olf_mitral = np.random.randn(n_mitral, n_glomeruli) * 0.1
        self.W_granule_mitral = np.random.randn(n_mitral, n_granule) * 0.05
        self.W_mitral_granule = np.random.randn(n_granule, n_mitral) * 0.05
        
        # Neuron properties
        self.tau_m = 20e-3   # Membrane time constant (s)
        self.V_rest = -65.0  # Resting potential (mV)
        self.V_thresh = -50.0  # Spike threshold (mV)
        
    def apply_neurogenesis(self, n_replace=None):
        """Replace a fraction of granule cells with new neurons."""
         n_replace  :
            n_replace = (.n_granule * .neurogenesis_rate)
        
        replace_idx = np.random.choice(.n_granule, n_replace, replace=)
        
        
        .W_granule_mitral[:, replace_idx] = np.random.randn(.n_mitral, n_replace) * 
        .W_mitral_granule[replace_idx, :] = np.random.randn(n_replace, .n_mitral) * 
        
         replace_idx
    
     ():
        
        n_steps = (duration / dt)
        
        V_mitral = np.ones(.n_mitral) * .V_rest
        V_granule = np.ones(.n_granule) * .V_rest
        
        spikes_mitral = np.zeros((n_steps, .n_mitral))
        
         t  (n_steps):
            
            I_ff = .W_olf_mitral @ odor_input
            
            
            I_inh = .W_granule_mitral @ (V_granule > .V_thresh).astype()
            
            
            dV_m = (-(V_mitral - .V_rest) + I_ff - I_inh) / .tau_m * dt
            V_mitral += dV_m
            
            
            fired = V_mitral >= .V_thresh
            spikes_mitral[t] = fired
            V_mitral[fired] = .V_rest
            
            
            I_exc = .W_mitral_granule @ fired.astype()
            dV_g = (-(V_granule - .V_rest) + I_exc) / .tau_m * dt
            V_granule += dV_g
        
         spikes_mitral
    
     ():
        
         np.mean(spikes[-(time_window / ):], axis=)


 :
    
    

     () -> :
        
        norm1, norm2 = np.linalg.norm(v1), np.linalg.norm(v2)
         norm1 ==   norm2 == :
             
         np.dot(v1, v2) / (norm1 * norm2)
    

     () -> np.ndarray:
        
        initial = representations[]
         np.array([
             - RepresentationalDriftAnalysis.cosine_similarity(initial, r)
             r  representations
        ])
    

     ():
        
        
        reps_naive = []
         _  ():
            model.apply_neurogenesis(neurogenesis_per_day)
            spikes = model.simulate_response(odor)
            reps_naive.append(model.population_vector(spikes))
        
        
        model2 = OlfactoryBulbModel()
        reps_experienced = []
         day  ():
            model2.apply_neurogenesis(neurogenesis_per_day)
            
             day %  == :  
                 _  (n_exposures):
                    spikes = model2.simulate_response(odor)
                    
                    pvec = model2.population_vector(spikes)
                    model2.W_olf_mitral +=  * np.outer(pvec, odor)
            
            spikes = model2.simulate_response(odor)
            reps_experienced.append(model2.population_vector(spikes))
        
        naive_drift = RepresentationalDriftAnalysis.measure_drift(reps_naive)
        exp_drift = RepresentationalDriftAnalysis.measure_drift(reps_experienced)
         naive_drift, exp_drift

When to Use

• Modeling adult neurogenesis effects in hippocampus, olfactory bulb, or cortex
• Studying representational drift in sensory systems
• Building computational models of learning-induced circuit changes
• Understanding how biological neural networks balance stability and plasticity
• Modeling olfactory system computations (odor discrimination, recognition)
• Continual learning in artificial neural networks inspired by neurogenesis

Key Insights for AI Systems

Biological Principle	AI Application
MOB population stability	Ensemble methods for stable feature representations
PCx representational drift	Temporal context encoding in sequence models
Experience-dependent protection	Rehearsal/replay in continual learning
Neurogenesis turnover	Growing neural networks, neuron dropout/replacement

Pitfalls

• Neurogenesis rate is species/region-specific (mice: ~1-2% granule cells/day in OB)
• New neuron integration time (weeks) must be modeled for accurate drift dynamics
• Model granularity (single neuron vs. population) affects stability predictions
• Experience-dependent stabilization requires realistic exposure statistics

Parameters

Parameter	Description	Biological Value
Neurogenesis rate	Fraction cells replaced/day	0.5–2% (OB granule)
Integration time	New neuron maturation	2–4 weeks
Drift timescale	Days to weeks for PCx drift	Weeks–months
Stabilization threshold	Exposures to protect representation	20–100

References

• Chen Z, Padmanabhan K. "Adult-neurogenesis allows for representational stability and flexibility in early olfactory system." eLife, 2026. DOI: 10.7554/eLife.107905
• Bhattacharya S, Bhattacharya S. "Olfactory bulb granule cells: New neurons in an old circuitry." Progress in Neurobiology, 2020.
• Rangel LM et al. "Temporally selective contextual encoding in the dentate gyrus of the hippocampus." Nature Communications, 2016.