Multiscale analysis of hybrid processes and reduction of stochastic neuron models. Gilles Wainrib joint work with: Khashayar Pakdaman and Michèle Thieullen Institut J.Monod- CNRS,Univ.Paris 6,Paris 7 - Labo. Proba et Modèles Aléatoires Univ.Paris 6,Paris 7,CNRS CREA Ecole polytechnique January, 2010 Part I : Introduction Deterministic neuron model Hodgkin Huxley (HH) model (Hodgkin Huxley - J.Physiol. 1952): Cm dV dt dm dt dh dt dn dt = I − gL (V − VL ) − gNa m3 h(V − VNa ) − gK n4 (V − VK ) = τm (V )−1 (m∞ (V ) − m) = τh (V )−1 (h∞ (V ) − h) = τn (V )−1 (n∞ (V ) − n) → Conductance-based neuron model Time-scale separation and reduction Sodium activation dynamic is faster than the other variables : τm → 0 m = m∞ (V ) Three-dimensional reduced system: Cm dV dt dh dt dn dt = I − gL (V − VL ) − gNa m∞ (V)3 h(V − VNa ) − gK n4 (V − VK ) = τh (V ) (h∞ (V ) − h) = τn (V ) (n∞ (V ) − n) Time-scale separation and reduction Sodium activation dynamic is faster than the other variables : τm → 0 m = m∞ (V ) Three-dimensional reduced system: Cm dV dt dh dt dn dt = I − gL (V − VL ) − gNa m∞ (V)3 h(V − VNa ) − gK n4 (V − VK ) = τh (V ) (h∞ (V ) − h) = τn (V ) (n∞ (V ) − n) Reduction of neuron models : key step in theoretical (singular perturbations) and numerical analysis Rinzel 1985, Kepler et al. 1992, Meunier 1992, Suckley et al.2003, Rubin et al. 2007, ... Modelling neurons with stochastic ion channels Single ion channels stochasticity: • Macromolecular devices : open and close through voltage-induced conformational changes Potassium channel • Stochasticity due to thermal noise Modelling neurons with stochastic ion channels Single ion channels stochasticity: • Macromolecular devices : open and close through voltage-induced conformational changes Potassium channel • Stochasticity due to thermal noise Channel noise : finite size effects responsible for intrinsic variability noise-induced phenomena (spontaneous activity, signal detection enhancement,...) Modelling neurons with stochastic ion channels Deterministic model X = (V , u) dV dt du dt = F (V , u) = (1 − u)α(V ) − uβ(V ) = τu (V )(u∞ (V ) − u) Modelling neurons with stochastic ion channels Stochastic model XN = (VN , uN ) Modelling neurons with stochastic ion channels Stochastic model XN = (VN , uN ) • Single ion channel i ∈ {1, ..., N} with voltage-dependent transition rates : independent jump Markov process ci (t) Modelling neurons with stochastic ion channels Stochastic model XN = (VN , uN ) • Single ion channel i ∈ {1, ..., N} with voltage-dependent transition rates : independent jump Markov process ci (t) • Proportion of open ion channels (empirical measure) :: N uN (t) = 1X ci (t) N i=1 Modelling neurons with stochastic ion channels Stochastic model XN = (VN , uN ) • Single ion channel i ∈ {1, ..., N} with voltage-dependent transition rates : independent jump Markov process ci (t) • Proportion of open ion channels (empirical measure) :: N uN (t) = 1X ci (t) N i=1 • Between the jumps, voltage dynamics: dVN = F (VN , uN ) dt Modelling neurons with stochastic ion channels Modelling neurons with stochastic ion channels • Modelling framework: Neuron ⇐⇒ population of globally coupled independent ion channels Modelling neurons with stochastic ion channels • Modelling framework: Neuron ⇐⇒ population of globally coupled independent ion channels • Mathematical framework: Piecewise-deterministic Markov process at the fluid limit Modelling neurons with stochastic ion channels • Modelling framework: Neuron ⇐⇒ population of globally coupled independent ion channels • Mathematical framework: Piecewise-deterministic Markov process at the fluid limit ⇓ (Davis, 1984) Modelling neurons with stochastic ion channels • Modelling framework: Neuron ⇐⇒ population of globally coupled independent ion channels • Mathematical framework: Piecewise-deterministic Markov process at the fluid limit ⇓ ⇓ (Kurtz, 1971) (Davis, 1984) Limit Theorems : Law of large numbers Theorem When N → ∞, XN converges to X in probability over finite time intervals [0, T ] Limit Theorems : Law of large numbers Theorem When N → ∞, XN converges to X in probability over finite time intervals [0, T ] For ∆ > 0, define " # PN (T , ∆) := P sup |XN (t) − X (t)|2 > ∆ t∈[0,T ] Then lim PN (T , ∆) = 0 N→∞ Limit Theorems : Law of large numbers Theorem When N → ∞, XN converges to X in probability over finite time intervals [0, T ] For ∆ > 0, define " # PN (T , ∆) := P sup |XN (t) − X (t)|2 > ∆ t∈[0,T ] Then lim PN (T , ∆) = 0 N→∞ More precisely, there exists constants B, C > 0 such that: lim sup N→∞ ∆e −BT 1 log PN (T , ∆) ≤ − N CT 2 Limit Theorems : Law of large numbers Theorem When N → ∞, XN converges to X in probability over finite time intervals [0, T ] For ∆ > 0, define " # PN (T , ∆) := P sup |XN (t) − X (t)|2 > ∆ t∈[0,T ] Then lim PN (T , ∆) = 0 N→∞ More precisely, there exists constants B, C > 0 such that: lim sup N→∞ ∆e −BT 1 log PN (T , ∆) ≤ − N CT 2 Pakdaman, Thieullen, W. ”Fluid limit theorems for stochastic hybrid systems with application to neuron models” (2009) arXiv:1001.2474 Limit Theorems : Central limit Theorem: Let √ RN (t) := „ N t Z XN (t) − « F (XN (s))ds 0 When N → ∞, RN converges in law to a diffusion process Z t R(t) = Σ(X (s))dWs 0 Limit Theorems : Central limit Theorem: Let √ RN (t) := „ N t Z XN (t) − « F (XN (s))ds 0 When N → ∞, RN converges in law to a diffusion process Z t R(t) = Σ(X (s))dWs 0 Langevin Approximation X̃N = (ṼN (t), ũN (t)): d ṼN (t) = F (ṼN (t), ũN (t))dt d ũN (t) = 1 b(ṼN (t), ũN (t))dt + √ Σ(ṼN (t), ũN (t))dWs N Limit Theorems : Central limit Theorem: Let √ RN (t) := „ N t Z XN (t) − « F (XN (s))ds 0 When N → ∞, RN converges in law to a diffusion process Z t R(t) = Σ(X (s))dWs 0 Langevin Approximation X̃N = (ṼN (t), ũN (t)): d ṼN (t) = F (ṼN (t), ũN (t))dt d ũN (t) = 1 b(ṼN (t), ũN (t))dt + √ Σ(ṼN (t), ũN (t))dWs N Further developments : strong approximation (pathwise CLT), Markov vs. Langevin, large deviations Stochastic reduction ? Part II : Mathematical analysis Singular perturbations for jump Markov processes Figure: Multiscale four-state model. Horizontal transitions are fast, whereas vertical transitions are slow. Singular perturbations for jump Markov processes Singular perturbations for jump Markov processes Singular perturbations for jump Markov processes : general setting Yin, Zhang, ”Continuous-time Markov Chains and Applications : a singular perturbation approach”, 1998 Assumption There exist n subsets of fast transitions. E = E1 ∪ E2 ∪ ... ∪ En • if i, j ∈ Ek then αi,j is of order O(−1 ), • otherwise, if i ∈ Ek and j ∈ El , with k 6= l then αi,j is of order O(1). Singular perturbations for jump Markov processes : general setting Constructing a reduced process: • quasi-stationary distributions (ρki )i∈Ek within fast subsets Ek , for k ∈ {1, ..., n}. • aggregated process (X̄ ) on the state space Ē = {1, ..., n} with transition rates: ᾱk,l = XX i∈Ek j∈El ρik αi,j for k, l ∈ Ē Singular perturbations for jump Markov processes : first-order Theorem • all-fast case For all t > 0, the probability Pi (t) = P [Xt = xi ] converges when → 0 to the stationary distribution ρi , for all i ∈ E . Singular perturbations for jump Markov processes : first-order Theorem • all-fast case For all t > 0, the probability Pi (t) = P [Xt = xi ] converges when → 0 to the stationary distribution ρi , for all i ∈ E . • multiscale case As → 0 the process (X ) is close to the reduced process (X̄ ). More precisely : 1. E hR T 0 “ ” i2 1{X (t)=xik } − ρki 1{X̄ =k} Φ(xik )dt = O(), for any function Φ : E → R, with k ∈ {1, ..., n} and i ∈ Ek . 2. The process X̄ converges in law to X̄ . Singular perturbations for jump Markov processes : second-order Rescaled process 1 n (t) = √ Theorem The rescaled process process Z T ` ´ 1{X (t)=xi } − ρi Φ(xi , s)ds 0 n (t) converges in law to the switching diffusion t Z n(t) = σ(s)dWs 0 where W is a standard n-dimensional Brownian motion. The diffusion matrix A = σ(s)σ 0 (s) is given by: ˆ ˜ Aij (s) = Φ(xi , s)Φ(xj , s) ρi R(i, j) + ρj R(j, i) where ∞ Z R(i, j) = 0 ` ´ P (i, j, t) − ρj dt Multiscale analysis of stochastic neuron models ) with Full model : XN = (VN , uN empirical measure for a population of multiscale jump processes • uN ) • V̇N = F (VN , uN Multiscale analysis of stochastic neuron models ) with Full model : XN = (VN , uN empirical measure for a population of multiscale jump processes • uN ) • V̇N = F (VN , uN Requires two extensions : 1. Population of jump processes 2. Piecewise deterministic Markov process Stationnary distribution for populations of multiscale jump processes Stationnary distributions for the empirical measure→ multinomial distributions Ex: two-state model k ρ(N) (k/N) = CNk u∞ (1 − u∞ )N−k Averaging method for PDMP Ex (all-fast): VN (t) = fast with uN t Z 0 F (VN (s), uN (s))ds Averaging method for PDMP Ex (all-fast): VN (t) = t Z 0 F (VN (s), uN (s))ds fast with uN Z → F̄N (V̄N ) := (N) F (V̄N , u)ρstat (du) (ergodic convergence) Averaging method for PDMP Ex (all-fast): VN (t) = t Z 0 F (VN (s), uN (s))ds fast with uN Z → F̄N (V̄N ) := (N) F (V̄N , u)ρstat (du) (ergodic convergence) ) converges in law • Theorem (general case) When → 0, the process (VN , uN towards a coarse-grained hybrid process: d V̄N = F̄N (V̄N , ūN ) dt and ū reduced jump process with averaged transition rates, functions of V̄ . Faggionato, Gabrielli, Ribezzi Crivellari 2009 Averaging method for PDMP Ex (all-fast): VN (t) = t Z 0 F (VN (s), uN (s))ds fast with uN Z → F̄N (V̄N ) := (N) F (V̄N , u)ρstat (du) (ergodic convergence) ) converges in law • Theorem (general case) When → 0, the process (VN , uN towards a coarse-grained hybrid process: d V̄N = F̄N (V̄N , ūN ) dt and ū reduced jump process with averaged transition rates, functions of V̄ . Faggionato, Gabrielli, Ribezzi Crivellari 2009 • Central limit theorem (ongoing work) → diffusion approximation : d ṼN dt = F̄N (ṼN , ũN )dt + √ σN (ṼN , ũN )dWt Part III : Application to Hodgkin-Huxley model Application Hodgkin-Huxley model : reduced model (two-state) Averaging ”m3 ” with respect to the binomial stationnary distribution (N) k (1 − m )N−k yields: ρm (k/N) = CNk m∞ ∞ Cm dV dt = I − gL (V − VL ) − gNa m∞ (V )3 h(V − VNa ) − gK n4 (V − VK ) − gNa h(V − VNa )KN (V) (supplementary terms) with KN (V ) = 3 1 m∞ (V )2 (1 − m∞ (V )) + 2 m∞ (V )(1 + 2m∞ (V )2 ) N N Important remark : Noise strength η := 1 N appears as a bifurcation parameter. Application Hodgkin-Huxley model : bifurcations of the reduced model Figure: Bifurcation diagram with η as parameter for I = 0 of system (HHN TS ). Application Hodgkin-Huxley model : bifurcations of the reduced model Figure: Two-parameter bifurcation diagram of system (HHN TS ) with I and η as parameters. Application Hodgkin-Huxley model : bifurcations of the reduced model 1. Below the double cycle curve is a region with a unique stable equilibrium point : ISI distribution should be approximately exponential, since a spike corresponds to a threshold crossing. Application Hodgkin-Huxley model : bifurcations of the reduced model 1. Below the double cycle curve is a region with a unique stable equilibrium point : ISI distribution should be approximately exponential, since a spike corresponds to a threshold crossing. 2. Between the double cycle and the Hopf curves is a bistable region : ISI distribution should be bimodal, one peak corresponding to the escape from the stable equilibrium, and the other peak to the fluctuations around the limit cycle. Application Hodgkin-Huxley model : bifurcations of the reduced model 1. Below the double cycle curve is a region with a unique stable equilibrium point : ISI distribution should be approximately exponential, since a spike corresponds to a threshold crossing. 2. Between the double cycle and the Hopf curves is a bistable region : ISI distribution should be bimodal, one peak corresponding to the escape from the stable equilibrium, and the other peak to the fluctuations around the limit cycle. 3. Above the Hopf curve is a region with a stable limit cycle and an unstable equilibrium point : ISI distribution should be centered around the period of the limit cycle. Application Hodgkin-Huxley model : stochastic simulations Figure: A. With N = 30 (zone 3), noisy periodic trajectory. B. With N = 70 (zone 2), bimodality of ISI’s C. With N = 120, ISI statistics are closer to a poissonian behavior. Application Hodgkin-Huxley model : stochastic simulations Figure: Interspike Interval (ISI) distributions Conclusions and perspectives • Systematic method for reducing a large class of stochastic neuron models • Based on recent mathematical developments of the averaging method • Illustration on HH : enables a bifurcation analysis with noise strength as parameter Conclusions and perspectives • Systematic method for reducing a large class of stochastic neuron models • Based on recent mathematical developments of the averaging method • Illustration on HH : enables a bifurcation analysis with noise strength as parameter • Other applications in neuroscience (synaptic models, networks, biochemical reactions) • Open mathematical questions (link with stochastic bifurcations, scaling in the double limit N → ∞, → 0) Singular perturbations for jump Markov processes : heuristics Law evolution : dP = dt „ Qs (t) + « 1 Qf (t) P with initial condition P (0) = p 0 . We are looking for an expansion of P (t) of the form Pr (t) = r X i=0 i φi (t) + r X i=0 t i ψi ( ) Singular perturbations for jump Markov processes : heuristics Identifying power of : Qf (t)φ0 (t) Qf (t)φ1 (t) = = ... Qf (t)φi (t) = 0 dφ0 (t) − φ0 (t)Qs (t) dt dφi−1 (t) − φi−1 (t)Qs (t) dt Error control: 1. |P (t) − Pr (t)| = O(r +1 ) uniformly in t ∈ [0, T ] 2. there exist K , k0 > 0 such that |ψi (t)| < Ke −k0 t Multiscale analysis of stochastic neuron models : summary Second order approximation for PDMP Central limit theorem 1 √ t „ Z Vt − 0 F (Vs )ds « t Z → σF (V̄s )dWs 0