王金光
(山東外貿(mào)職業(yè)學(xué)院,山東 青島 266100)
突觸信號(hào)傳導(dǎo)的動(dòng)態(tài)飽和模型研究
王金光
(山東外貿(mào)職業(yè)學(xué)院,山東 青島 266100)
對(duì)一類動(dòng)態(tài)飽和突觸神經(jīng)模型中信號(hào)傳導(dǎo)性質(zhì)進(jìn)行了研究。模型的動(dòng)態(tài)過(guò)程采用高階Milstein隨機(jī)微分方程解法進(jìn)行求解,其信號(hào)輸入輸出特性用集平均互相關(guān)系數(shù)進(jìn)行衡量。集平均互相關(guān)系數(shù)的數(shù)值分析結(jié)果表明,適宜的噪聲能夠增強(qiáng)信號(hào)傳導(dǎo),并且通過(guò)調(diào)節(jié)飽和勢(shì)比值大小和突觸神經(jīng)群體數(shù)目,觀測(cè)到噪聲增強(qiáng)信號(hào)傳導(dǎo)的非線性現(xiàn)象更加顯著。
突觸神經(jīng)模型;互相關(guān)系數(shù);信號(hào)傳導(dǎo);噪聲增強(qiáng)
近30年的研究表明,噪聲能夠在非線性系統(tǒng)的信號(hào)傳輸中起到協(xié)同作用,這類非線性現(xiàn)象稱為隨機(jī)共振現(xiàn)象或噪聲增強(qiáng)現(xiàn)象[1-7]。神經(jīng)細(xì)胞中信號(hào)傳導(dǎo)的非線性動(dòng)態(tài)過(guò)程發(fā)生在噪聲環(huán)境中,Longtin等[2]首次在FitzHugh神經(jīng)元模型中實(shí)現(xiàn)了噪聲增強(qiáng)現(xiàn)象。隨后在Hodgkin-Huxley[3]、Hindmarsh-Rose[1]等神經(jīng)元模型中均發(fā)現(xiàn)了噪聲增強(qiáng)的信息處理現(xiàn)象。由于神經(jīng)信號(hào)的非周期性質(zhì),Collins等[4]提出了非周期隨機(jī)共振理論,這一理論的提出對(duì)于理解神經(jīng)細(xì)胞、中樞神經(jīng)乃至人腦的信息處理提供了一種新的思路。同時(shí),非周期隨機(jī)共振理論在人體平衡性[5]和人工耳蝸[6]等醫(yī)學(xué)工程方面的實(shí)際應(yīng)用發(fā)展迅速。中國(guó)國(guó)內(nèi)學(xué)者對(duì)于各種神經(jīng)元中的隨機(jī)共振現(xiàn)象及其應(yīng)用也進(jìn)行了大量研究[8-13],在語(yǔ)音處理[8]、神經(jīng)網(wǎng)絡(luò)功能[11-12]和圖像復(fù)原[13]等方面取得了很多重要研究成果。
文獻(xiàn)[1]~[6],[8]~[13]主要分析了噪聲協(xié)助弱信號(hào)克服細(xì)胞勢(shì)電位發(fā)放的閾值,以達(dá)到提高信息傳導(dǎo)效率的目的。但是,在突觸神經(jīng)信號(hào)傳導(dǎo)水平中,還有一類飽和動(dòng)態(tài)過(guò)程[7],傳導(dǎo)信號(hào)引發(fā)了突觸間隙囊泡神經(jīng)遞質(zhì)的釋放。但是,由于突觸囊泡群體的有限性,遞質(zhì)的釋放活動(dòng)具有一個(gè)飽和值,其信息傳遞過(guò)程可以用一類飽和突觸模型[7]來(lái)描述。相對(duì)于傳導(dǎo)信號(hào)來(lái)講,突觸神經(jīng)細(xì)胞內(nèi)外離子的隨機(jī)活動(dòng)可以視為白噪聲。這些隨機(jī)噪聲對(duì)于飽和突觸模型信號(hào)傳導(dǎo)的影響值得深入研究。
本文主要針對(duì)一類飽和突觸模型的信號(hào)傳導(dǎo)特性進(jìn)行了深入研究,首先利用改進(jìn)的高階Milstein解法[14]對(duì)模型所滿足的隨機(jī)微分方程進(jìn)行了求解。用集平均互相關(guān)系數(shù)這個(gè)衡量指標(biāo)對(duì)模型的信號(hào)輸入輸出特性進(jìn)行了深入分析。集平均互相關(guān)系數(shù)的數(shù)值分析結(jié)果表明,適宜的噪聲能夠增強(qiáng)信號(hào)傳導(dǎo)。通過(guò)調(diào)節(jié)飽和勢(shì)比值的大小和勢(shì)和興奮性(抑制性)突觸神經(jīng)群體數(shù)目,發(fā)現(xiàn)噪聲增強(qiáng)信號(hào)傳導(dǎo)的現(xiàn)象更加顯著。這些研究結(jié)果對(duì)于理解突觸信號(hào)處理機(jī)制具有重要意義。
突觸間隙囊泡神經(jīng)遞質(zhì)活動(dòng)的飽和動(dòng)態(tài)模型為[7]
這里,τ>0為松弛時(shí)間常數(shù),α>0為輸入轉(zhuǎn)化為輸出的參數(shù),Xs>0為飽和勢(shì),s(t)為輸入信號(hào)。非負(fù)噪聲ξ(t)為伽馬噪聲,其分布為
由于式(1)中系數(shù)含有隨機(jī)項(xiàng),這里采用基于伊藤-泰勒展開的高階Milstein隨機(jī)微分方程解法[14]。將式(1)寫為
這里,A(x(t))=-x(t)+α(Xs-x(t))s(t),B(x(t))=α(Xs-x(t)),dψ(t)=ξ(t)dt。數(shù)值求解時(shí),將時(shí)間進(jìn)行離散化,采樣時(shí)間Δt=ti+1-ti,i=0,1,2,…,那么Δψi=ψ(ti+1)-ψ(ti)。由初始值x(t0),過(guò)程x(t)在時(shí)刻ti+1的解為
這里,B′(x(t))=-α。式(4)是由依據(jù)伊藤-泰勒展開的 Milstein迭代隨機(jī)微分方程解,其收斂階數(shù)比Euler-Maruyama解法高[14]。
為衡量非周期信號(hào)s(t)在此模型的傳輸特性,計(jì)算輸入信號(hào)s(t)和系統(tǒng)輸出x(t)的互相關(guān)系數(shù)
這里,〈·〉表示時(shí)間平均算子。實(shí)驗(yàn)中,互相關(guān)系數(shù)對(duì)于相同強(qiáng)度的不同噪聲樣本進(jìn)行集平均,得到集平均互相關(guān)系數(shù)E[Csx]。
進(jìn)一步,考慮抑制性突觸飽和神經(jīng)元對(duì)于信號(hào)傳輸?shù)挠绊憽,F(xiàn)實(shí)中存在興奮性和抑制性兩種突觸神經(jīng)元,當(dāng)式(1)中飽和勢(shì)Xs>0時(shí),突觸神經(jīng)是興奮性的,當(dāng)飽和勢(shì)Xs變?yōu)閄I(XI<0),突觸神經(jīng)是抑制性的,即
這里,抑制性突觸飽和神經(jīng)元釋放抑制性遞質(zhì),由于其離子通道動(dòng)力學(xué)性質(zhì)[15],可以使他們的突觸后神經(jīng)元被抑制,這里僅考慮由噪聲驅(qū)動(dòng)。興奮性突觸神經(jīng)元和抑制性突觸神經(jīng)元各選取1 000個(gè),飽和勢(shì)比值XI/Xs為-5/7、-1和-2,參數(shù)α=100。圖2給出了集平均互相關(guān)系數(shù)E[Cxs]隨著噪聲強(qiáng)度(均方根r)的變化曲線??梢钥闯?,抑制性突觸神經(jīng)元雖然只有噪聲的驅(qū)動(dòng),但是在大量的抑制性突觸神經(jīng)元與興奮性突觸神經(jīng)元組成的多信號(hào)傳輸通道中,集平均互相關(guān)系數(shù)E[Csx]依然對(duì)應(yīng)了一個(gè)最優(yōu)的噪聲強(qiáng)度。并且,飽和勢(shì)比值XI/Xs對(duì)于集平均互相關(guān)系數(shù)的影響較為明顯,特別是XI/Xs=-1時(shí),集平均互相關(guān)系數(shù)E[Csx]的最大值達(dá)到0.92,這是一種非常適合信號(hào)傳輸?shù)耐挥|神經(jīng)元群體,這一結(jié)果對(duì)于理解突觸信號(hào)處理機(jī)制具有重要意義。比如,如何優(yōu)化飽和勢(shì)比值XI/Xs來(lái)增強(qiáng)突觸信號(hào)傳導(dǎo)是值得進(jìn)一步研究的問題,而且,現(xiàn)實(shí)中興奮性突觸神經(jīng)元和抑制性突觸神經(jīng)元是否選擇類似的優(yōu)化策略來(lái)處理突觸信號(hào),更是值得探討的研究方向。
圖1 集平均互相關(guān)系數(shù)E[Csx]隨著噪聲強(qiáng)度(均方根r)的變化Fig.1 Ensemble-averaged correlation coefficient as a function of the noise rms amplitude
圖2 集平均互相關(guān)系數(shù)E[Csx]隨著噪聲強(qiáng)度(均方根r)的變化Fig.2 Ensemble-averaged correlation coefficient as a function of the noise rms amplitude
本文針對(duì)一類飽和突觸模型的信號(hào)傳導(dǎo)特性進(jìn)行了研究,利用Milstein解法對(duì)模型所滿足的隨機(jī)微分方程進(jìn)行了求解。用集平均互相關(guān)系數(shù)這個(gè)衡量指標(biāo)對(duì)模型的信號(hào)輸入輸出特性進(jìn)行了分析。集平均互相關(guān)系數(shù)的數(shù)值分析結(jié)果表明,適宜的噪聲能夠增強(qiáng)信號(hào)傳導(dǎo)。同時(shí),研究了大量的抑制性突觸神經(jīng)元與興奮性突觸神經(jīng)元組成的多信號(hào)傳輸通道中噪聲增強(qiáng)信號(hào)傳導(dǎo)的現(xiàn)象,分析結(jié)果表明噪聲強(qiáng)度能夠優(yōu)化集平均互相關(guān)系數(shù),增強(qiáng)信號(hào)傳導(dǎo)。不同的飽和勢(shì)比值對(duì)于集平均互相關(guān)系數(shù)的影響非常顯著,因此飽和勢(shì)比值的優(yōu)化非常值得進(jìn)一步深入研究。
[1]Gammaitoni L,Hanggi P,Jung P,et al.Stochastic resonance[J].Reviews of Modern Physics,1998,70(1):233-287.
[2]Longtin A.Stochastic resonance in neuron models[J].Journal of Statistical Physics,1993,70(1-2):309-327.
[3]Gong Y,Xie Y,Hao Y.Coherence resonance induced by non-Gaussian noise in a deterministic Hodgkin Huxley neuron[J].Physica A,2009,388(18):3759-3764.
[4]Collins J J,Chow C C,Capela A C,et al.Aperiodic stochastic resonance[J].Physical Review E,1996,54(5):5575-5584.
[5]Dhruv N T,Niemi J B,Harry J D,et al.Enhancing tactile sensation in older adults with electrical noise stimulation[J].Neuroreport,2002,13(5):597-600.
[6]Morse R P,Evans E F.Enhancement of vowel coding for cochlear implants by addition of noise[J].Nature Medicine,1996,2(8):928-932.
[7]Chapeau-Blondeau F,Duan F,Abbott D.Synaptic signal transduction aided by noise in a dynamical saturating model[J].Physical Review E,2010,81(2):021124.
[8]薛凌云,向?qū)W勤,范影樂,等.基于神經(jīng)元閾上非周期隨機(jī)共振機(jī)制的語(yǔ)音復(fù)原技術(shù)研究[J].傳感技術(shù)學(xué)報(bào),2009,22(2):213-218.
Xue Linyun,Xiang Xueqin,F(xiàn)an Yingle,et al.Speech restoration based on suprathreshold stochastic resonance of neuron model[J].Chinese Journal of Sensors and Actuators,2009,22(2):213-218.
[9]李冰,彭建華,劉延柱.隨機(jī)延時(shí) Hodgkin-Huxley神經(jīng)網(wǎng)絡(luò)的同步與聯(lián)想記憶[J].上海交通大學(xué)學(xué)報(bào),2005,39(11):1924-1928.
Li Bing,Peng Jianhua,Liu Yanzhu.The synchronization and associative memory of Hodgkin-Huxley neural network with randomly distributed time delays[J].Journal of Shanghai Jiaotong University,2005,39(11):1924-1928.
[10]王俊琦,焦賢發(fā).基于IF模型閾值神經(jīng)元的隨機(jī)共振[J].合肥工業(yè)大學(xué)學(xué)報(bào),2010,33(6):939-943.
Wang Junqi,Jiao Xianfa.Stochastic resonance of an integrate-and-fire neuron model with threshold[J].Journal of Hefei University of Technology,2010,33(6):939-943.
[11]曹明明,王景,段法兵.興奮性神經(jīng)群體的非周期信息傳輸研究[J],復(fù)雜系統(tǒng)與復(fù)雜性科學(xué),2009,6(4):66-70.
Cao Mingming,Wang Jing,Duan Fabin.Study of aperiodic information transmission in ensembles of excitable neurons[J].Complex Systems and Complexity Science,2009,6(4):66-70.
[12]于海濤,王江,劉晨,等.耦合小世界神經(jīng)網(wǎng)絡(luò)的隨機(jī)共振[J],物理學(xué)報(bào),2012,61(6):068702.
Yu Haitao,Wang Jiang,Liu Chen,et al.Stochastic resonance in coupled small-world neural networks[J].Acta Physica Sinica,2012,61(6):068702.
[13]龔振宇,龐全,范影樂.自適應(yīng)隨機(jī)共振的圖像復(fù)原研究[J].計(jì)算機(jī)工程與科學(xué),2009,31(5):46-49.
Gong Zhenyu,Pang Quan,F(xiàn)an Yingle.The image restoration research by auto-adapted stochastic resonating[J].Computer Engineering &Science,2009,31(5):46-49.
[14]Higham D J.An algorithmic introduction to numerical simulation of stochastic differential equations[J].SIAM Review,2001,43(3):525-546.
[15]Chapeau-Blondeau F,Chambet N.Synapse models for neural networks:from ion channel kinetics to multiplicative coefficient[J].Neural Computation,1995,7(2):713-734.
On Synaptic Signal Transduction in a Dynamical Saturating Model
WANG Jin-guang
(Shandong Foreign Trade Vocational College,Qingdao 266100,China)
The synaptic signal transduction in a dynamical saturation neuron model is studied.At the pre-synaptic and post-synaptic stages,the evolution of neurotransmitter molecules in the synaptic cleft can be described by a dynamical saturation model.In the presence of noise,the signal transduction in this model is characterized by the ensemble-averaged correlation coefficient.The evolution of synaptic signal transmission is solved by the Milstein's high-order method of stochastic differential equation.The numerical result of the ensemble average correlation coefficient demonstrates the effect of noise-enhanced signal transduction in a single neuron model and an ensemble population of synaptic saturation neurons.Moreover,the noise-enhanced signal transduction effect is more visible by tuning the ratio of saturating current and the population of neurons.
neural synaptic model;correlation coefficient;signal transduction;noise enhancement
Q612;N945.12
A
1672-3813(2013)02-0059-04
2012-12-25
山東省自然科學(xué)基金(ZR2010FM006)
王金光(1976-),男,山東寧津人,碩士,講師,主要研究方向?yàn)橄到y(tǒng)理論。
(責(zé)任編輯 耿金花)