// striatum model solved using mid-point method (aka 2nd order Runge-Kutta)
// this version a Win32-compatible re-write (17/5/2010) of code from March 2009
#include <mex.h>
#include <matrix.h>
#include <stdio.h>
#include <math.h>
#include <stdlib.h>
#include <time.h>
// MEX-specific macros for isnan etc as Microsoft VS etc do not support them
// from: http://www.mathworks.com/matlabcentral/newsreader/view_thread/130557
#if __STDC_VERSION_ <= 199901L
# define isinf(a) ((a) == mxGetInf() || (a) == -mxGetInf())
# define isnan(a) ((a) != (a))
# define isfinite(a) !((a) != (a) || (a) == mxGetInf() || (a) == -mxGetInf())
#endif
// ===============================================================
// RANDOM NUMBER GENERATOR
#define SHR3 (jz=jsr, jsr^=(jsr<<13), jsr^=(jsr>>17), jsr^=(jsr<<5),jz+jsr)
#define UNI (.5 + (signed) SHR3 * .2328306e-9)
#define RNOR (hz=SHR3, iz=hz&127, (abs(hz)<kn[iz])? hz*wn[iz] : nfix())
#define REXP (jz=SHR3, iz=jz&255, ( jz <ke[iz])? jz*we[iz] : efix())
static unsigned int iz,jz,jsr=123456789,kn[128],ke[256];
static int hz; static float wn[128],fn[128], we[256],fe[256];
float nfix(void) { /*provides RNOR if #define cannot */
const float r = 3.442620f; static float x, y;
for(;;){ x=hz*wn[iz];
if(iz==0){ do{x=-log(UNI)*0.2904764; y=-log(UNI);} while(y+y<x*x);
return (hz>0)? r+x : -r-x;
}
if( fn[iz]+UNI*(fn[iz-1]-fn[iz]) < exp(-.5*x*x) ) return x;
hz=SHR3; iz=hz&127;if(abs(hz)<kn[iz]) return (hz*wn[iz]);
}
}
float efix(void) { /*provides REXP if #define cannot */
float x;
for(;;){
if(iz==0) return (7.69711-log(UNI));
x=jz*we[iz];
if( fe[iz]+UNI*(fe[iz-1]-fe[iz]) < exp(-x) ) return (x);
jz=SHR3; iz=(jz&255);
if(jz<ke[iz]) return (jz*we[iz]);
}
}
// == This procedure sets the seed and creates the tables ==
void zigset(unsigned int jsrseed) {
const double m1 = 2147483648.0, m2 = 4294967296.;
double dn=3.442619855899,tn=dn,vn=9.91256303526217e-3, q;
double de=7.697117470131487, te=de, ve=3.949659822581572e-3;
int i; jsr=jsrseed;
/* Tables for RNOR: */
q=vn/exp(-.5*dn*dn);
kn[0]=(dn/q)*m1; kn[1]=0;
wn[0]=q/m1; wn[127]=dn/m1;
fn[0]=1.; fn[127]=exp(-.5*dn*dn);
for(i=126;i>=1;i--) {
dn=sqrt(-2.*log(vn/dn+exp(-.5*dn*dn)));
kn[i+1]=(dn/tn)*m1; tn=dn;
fn[i]=exp(-.5*dn*dn); wn[i]=dn/m1;
}
/* Tables for REXP */
q = ve/exp(-de);
ke[0]=(de/q)*m2; ke[1]=0;
we[0]=q/m2; we[255]=de/m2;
fe[0]=1.; fe[255]=exp(-de);
for(i=254;i>=1;i--) {
de=-log(ve/de+exp(-de));
ke[i+1]= (de/te)*m2; te=de;
fe[i]=exp(-de); we[i]=de/m2;
}
}
int mxGetScalarInt32(const mxArray* a, int defaultValue = -2147483648)
{
if (mxIsEmpty(a))
{
if (defaultValue == -2147483648) throw "missing input";
return defaultValue;
}
if (!mxIsInt32(a)) throw "not int32";
if (mxGetNumberOfDimensions(a) != 2) throw "expected scalar";
if (mxGetM(a) != 1 || mxGetN(a) != 1) throw "expected scalar";
return mxGetScalar(a);
//if (fabs(val) > pow(2,30)) throw "value out of range";
//if (floor(val) != val) throw "value not an integer";
//return val;
}
double mxGetScalarDouble(const mxArray* a, double defaultValue = -2147483648.1)
{
if (mxIsEmpty(a))
{
if (defaultValue == -2147483648.1) throw "missing input";
return defaultValue;
}
if (!mxIsDouble(a)) throw "not Double";
if (mxGetNumberOfDimensions(a) != 2) throw "expected scalar";
if (mxGetM(a) != 1 || mxGetN(a) != 1) throw "expected scalar";
return mxGetScalar(a);
}
struct MatrixInt32
{
int* data;
int M, N;
};
MatrixInt32 mxGetMatrixInt32(const mxArray* a, int M = -1, int N = -1)
{
MatrixInt32 ret;
if (!mxIsInt32(a)) throw "not int32";
if (mxGetNumberOfDimensions(a) != 2) throw "expected 2D matrix";
if (mxIsComplex(a)) throw "expected real matrix";
if (mxIsEmpty(a)) ret.data = NULL;
else ret.data = (int*)mxGetData(a);
ret.M = mxGetM(a);
ret.N = mxGetN(a);
if (M != -1 && M != ret.M) throw "interger matrix has wrong dimension (M)";
if (N != -1 && N != ret.N) throw "interger matrix has wrong dimension (N)";
return ret;
}
struct MatrixDouble
{
double* data;
int M, N;
};
MatrixDouble mxGetMatrixDouble(const mxArray* a, int M = -1, int N = -1)
{
MatrixDouble ret;
if (!mxIsDouble(a)) throw "not double";
if (mxGetNumberOfDimensions(a) != 2) throw "expected 2D matrix";
if (mxIsComplex(a)) throw "expected real matrix";
if (mxIsEmpty(a)) ret.data = NULL;
else ret.data = (double*)mxGetData(a);
ret.M = mxGetM(a);
ret.N = mxGetN(a);
if (M != -1 && M != ret.M) throw "double matrix has wrong dimension (M)";
if (N != -1 && N != ret.N) throw "double matrix has wrong dimension (N)";
return ret;
}
// ===============================================================
// Main simulation function
void execute(int nlhs, mxArray *plhs[],
int nrhs, const mxArray *prhs[])
{
// ===============================================================
// The Inputs
// ---------------------------------------------------------------
if (nrhs != 72) throw "not enough inputs";
// ---------------------------------------------------------------
// Get the inputs
double tstart = mxGetScalarDouble(prhs[0]);
double tfinal = mxGetScalarDouble(prhs[1]);
double dt = mxGetScalarDouble(prhs[2]);
// double dt_fs = mxGetScalarDouble(prhs[3]);
int tend = (int)(tfinal / dt);
MatrixDouble MSparams = mxGetMatrixDouble(prhs[3]);
MatrixDouble FSparams = mxGetMatrixDouble(prhs[4]);
double Eglu = mxGetScalarDouble(prhs[5]);
double Egaba = mxGetScalarDouble(prhs[6]);
double ts_glu_AMPA = mxGetScalarDouble(prhs[7]);
double ts_glu_NMDA = mxGetScalarDouble(prhs[8]);
double ts_gaba = mxGetScalarDouble(prhs[9]);
double tau_fsgap = mxGetScalarDouble(prhs[10]);
int MSspikebuffer = mxGetScalarInt32(prhs[11]);
int FSspikebuffer = mxGetScalarInt32(prhs[12]);
MatrixDouble initVms = mxGetMatrixDouble(prhs[13]);
MatrixDouble initUms = mxGetMatrixDouble(prhs[14], initVms.M, initVms.N);
MatrixDouble initVfs = mxGetMatrixDouble(prhs[15]);
MatrixDouble initUfs = mxGetMatrixDouble(prhs[16], initVfs.M, initVfs.N);
MatrixDouble initVfsgap = mxGetMatrixDouble(prhs[17]);
// get the dimensions of the MS and FS networks
const int *dims_ms = mxGetDimensions(prhs[13]);
int ndim_ms = mxGetNumberOfDimensions(prhs[13]);
int N_MS = initVms.M;
const int *dims_fs = mxGetDimensions(prhs[15]);
int ndim_fs = mxGetNumberOfDimensions(prhs[15]);
int N_FS = initVfs.M;
const int *dims_fsgap = mxGetDimensions(prhs[17]);
int ndim_fsgap = mxGetNumberOfDimensions(prhs[17]);
int N_fsgap = initVfsgap.M*initVfsgap.N;
MatrixDouble initSEQ_MSglu = mxGetMatrixDouble(prhs[18], N_MS);
MatrixDouble initSEQ_FSglu = mxGetMatrixDouble(prhs[19], N_FS);
MatrixDouble initSEQ_MSGABA = mxGetMatrixDouble(prhs[20], N_MS);
MatrixDouble initSEQ_FSGABA = mxGetMatrixDouble(prhs[21], N_FS);
MatrixDouble initCTX = mxGetMatrixDouble(prhs[22]);
// *** NEED TO REMOVE THIS AND FIND ANOTHER WAY TO SET UP MATRICES FURTHER DOWN ***
const int *initSEQ_MSGABA_dims = mxGetDimensions(prhs[20]);
int initSEQ_MSGABA_ndim = mxGetNumberOfDimensions(prhs[20]);
const int *initSEQ_FSGABA_dims = mxGetDimensions(prhs[21]);
int initSEQ_FSGABA_ndim = mxGetNumberOfDimensions(prhs[21]);
// *** Not used currently, but will need bounds check put on dimensions ***
MatrixDouble Ims = mxGetMatrixDouble(prhs[23]);
MatrixDouble Ifs = mxGetMatrixDouble(prhs[24]);
MatrixInt32 Cctms = mxGetMatrixInt32(prhs[25], N_MS, 1); // need to set bound on to N_MS, no + 1
MatrixInt32 Cctms_b = mxGetMatrixInt32(prhs[26], N_MS+1, 1);
MatrixInt32 Cctms_d = mxGetMatrixInt32(prhs[27], N_MS, 1);
MatrixDouble Cctms_w = mxGetMatrixDouble(prhs[28], N_MS, 1);
// double a_ms = mxGetScalarDouble(prhs[29]);
MatrixInt32 Cmsms = mxGetMatrixInt32(prhs[29]);
MatrixInt32 Cmsms_b = mxGetMatrixInt32(prhs[30], N_MS+1, 1);
MatrixInt32 Cmsms_d = mxGetMatrixInt32(prhs[31], Cmsms.M, Cmsms.N);
MatrixDouble Cmsms_w = mxGetMatrixDouble(prhs[32], Cmsms.M, Cmsms.N);
MatrixInt32 Cfsms = mxGetMatrixInt32(prhs[33]);
MatrixInt32 Cfsms_b = mxGetMatrixInt32(prhs[34], N_FS+1, 1);
MatrixInt32 Cfsms_d = mxGetMatrixInt32(prhs[35], Cfsms.M, Cfsms.N);
MatrixDouble Cfsms_w = mxGetMatrixDouble(prhs[36], Cfsms.M, Cfsms.N);
const int *dims_Cfsms_b = mxGetDimensions(prhs[34]);
MatrixInt32 Cctfs = mxGetMatrixInt32(prhs[37], N_FS, 1);
MatrixInt32 Cctfs_b = mxGetMatrixInt32(prhs[38], N_FS+1, 1);
MatrixInt32 Cctfs_d = mxGetMatrixInt32(prhs[39], N_FS, 1);
MatrixDouble Cctfs_w = mxGetMatrixDouble(prhs[40], N_FS, 1);
const int *dims_Cctfs_b = mxGetDimensions(prhs[38]);
// double a_fs = mxGetScalarDouble(prhs[42]);
MatrixInt32 Cfsfs = mxGetMatrixInt32(prhs[41]);
MatrixInt32 Cfsfs_b = mxGetMatrixInt32(prhs[42], N_FS+1, 1);
MatrixInt32 Cfsfs_d = mxGetMatrixInt32(prhs[43], Cfsfs.M, Cfsfs.N);
MatrixDouble Cfsfs_w = mxGetMatrixDouble(prhs[44], Cfsfs.M, Cfsfs.N);
const int *dims_Cfsfs_b = mxGetDimensions(prhs[42]);
MatrixInt32 Cgapfs = mxGetMatrixInt32(prhs[45]);
MatrixInt32 Cgapfs_b = mxGetMatrixInt32(prhs[46], N_FS+1, 1);
MatrixDouble Cgapfs_w = mxGetMatrixDouble(prhs[47], Cgapfs.M, Cgapfs.N);
MatrixInt32 Pgapfs = mxGetMatrixInt32(prhs[48]);
MatrixDouble CTX_state = mxGetMatrixDouble(prhs[49]);
MatrixInt32 CHAN1_MS = mxGetMatrixInt32(prhs[50]);
MatrixInt32 CHAN1_FS = mxGetMatrixInt32(prhs[51]);
MatrixInt32 CHAN2_MS = mxGetMatrixInt32(prhs[52]);
MatrixInt32 CHAN2_FS = mxGetMatrixInt32(prhs[53]);
const int *dims_CHAN1_MS = mxGetDimensions(prhs[50]);
int N_CHAN1_MS = dims_CHAN1_MS[0];
const int *dims_CHAN1_FS = mxGetDimensions(prhs[51]);
int N_CHAN1_FS = dims_CHAN1_FS[0];
MatrixInt32 N_MSSEG = mxGetMatrixInt32(prhs[54], N_MS, 1);
MatrixDouble r_MSSEG = mxGetMatrixDouble(prhs[55], N_MS, 1);
MatrixDouble alpha_MSSEG = mxGetMatrixDouble(prhs[56], N_MS, 1);
MatrixInt32 N_FSSEG = mxGetMatrixInt32(prhs[57], N_FS, 1);
MatrixDouble r_FSSEG = mxGetMatrixDouble(prhs[58], N_FS, 1);
MatrixDouble alpha_FSSEG = mxGetMatrixDouble(prhs[59], N_FS, 1);
double glu_ratio = mxGetScalarDouble(prhs[60]);
double DA = mxGetScalarDouble(prhs[61]);
MatrixInt32 RecordChan_MS = mxGetMatrixInt32(prhs[62]);
const int dims_RecordChan_MS_out [2] = {tend, RecordChan_MS.M};
MatrixDouble PULSE = mxGetMatrixDouble(prhs[63]);
double Nctx_ms = mxGetScalarDouble(prhs[64]);
double Nctx_fs = mxGetScalarDouble(prhs[65]);
double ts_spks = mxGetScalarDouble(prhs[66]);
MatrixInt32 Pt = mxGetMatrixInt32(prhs[67]);
MatrixInt32 Pch = mxGetMatrixInt32(prhs[68], Pt.M, Pt.N);
MatrixDouble Phz = mxGetMatrixDouble(prhs[69], Pt.M, Pt.N);
int RANDSEED = mxGetScalarInt32(prhs[70]);
const char* filename = mxArrayToString(prhs[71]);
// ===============================================================
// set the random numer seeds
zigset(RANDSEED);
// ===============================================================
// Set up easy access to the MS and FS parameters
#define _MS(__m,__n) (MSparams.data[__n * MSparams.M + __m])
#define _FS(__m,__n) (FSparams.data[__n * FSparams.M + __m])
// ===============================================================
// get some dimensions
if (!filename) throw "no filename supplied for log file";
// open log file
FILE* fid = fopen(filename, "w");
if (!fid) throw "failed open log file";
fprintf(fid, "*************************************************** \n");
fprintf(fid, "*************** STARTING SIMULATION *************** \n");
fprintf(fid, "*************************************************** \n");
fprintf(fid, "tstart = %f msec \n", tstart);
fprintf(fid, "tfinal = %f msec \n", tfinal);
fprintf(fid, "dt = %f msec \n", dt);
fprintf(fid, "tend = %i iterations \n", tend);
fprintf(fid, " \n");
fprintf(fid, "N_MS = %i \n", N_MS);
fprintf(fid, "Example MS parameters are C = %f, vr = %f, vt = %f, a = %f, b = %f, c = %f, d = %f, vp = %f, k = %f, EDA = %f, alpha = %f, beta1 = %f, beta2 = %f, gDA = %f \n", _MS(0,0), _MS(0,1), _MS(0,2), _MS(0,3), _MS(0,4), _MS(0,5), _MS(0,6), _MS(0,7), _MS(0,8), _MS(0,9), _MS(0,10), _MS(0,11), _MS(0,12), _MS(0,13), _MS(0,14));
fprintf(fid, " \n");
fprintf(fid, "N_FS = %i \n", N_FS);
fprintf(fid, "Example FS parameters are C = %f, vr = %f, vt = %f, k = %f, a = %f, b = %f, c = %f, d = %f, vpeak = %f, vb = %f, eta = %f, epsilon = %f \n", _FS(0,0), _MS(0,1), _FS(0,2), _FS(0,3), _FS(0,4), _FS(0,5), _FS(0,6), _FS(0,7), _FS(0,8), _FS(0,9), _FS(0,10), _FS(0,11));
fprintf(fid, " \n");
fprintf(fid, "Eglu = %f \n", Eglu);
fprintf(fid, "Egaba = %f \n", Egaba);
fprintf(fid, "ts_glu_AMPA = %f \n", ts_glu_AMPA);
fprintf(fid, "ts_glu_NMDA = %f \n", ts_glu_NMDA);
fprintf(fid, "ts_gaba = %f \n", ts_gaba);
fprintf(fid, "tau_fsgap = %f \n", tau_fsgap);
fprintf(fid, " \n");
fprintf(fid, "gAMPA_MS = %f \n", Cctms_w.data[0]);
fprintf(fid, "gNMDA_MS = %f \n", glu_ratio * Cctms_w.data[0]);
fprintf(fid, "gGLUTAMATE_FS = %f \n", Cctfs_w.data[0]);
fprintf(fid, " \n");
fprintf(fid, "MSspikebuffer = %i \n", MSspikebuffer);
fprintf(fid, "FSspikebuffer = %i \n", FSspikebuffer);
fprintf(fid, " \n");
fprintf(fid, "*************************************************** \n");
fclose(fid);
// ===============================================================
// Set and init some parameters
double T; // Simulation Time
// int extasteps = (int)(dt_ms / dt_fs); // Number of extra iterations needed to solve the FS neurons
float VMS,UMS,VFS,UFS,U; // temp variables for the ms and fs neurons
float Vms1,Ums1,Vfs1,Ufs1,tmp; // temp variables for mid-point
float hdt = dt*0.5; // half time-step for midpoint method
// parameters for the MS neurons
int MSspikecount = 0;
int SEQind_ms = 0;
int SpikeEventCycle_MSGABA = 0;
int maxSEQdelay_MSGABA = initSEQ_MSGABA.N - 1;
int SpikeQueSize_MS = N_MS * (maxSEQdelay_MSGABA+1);
double *Mg_ms; Mg_ms = (double *)malloc(N_MS * sizeof(double)); // Mg block for the MS NMDA channels
double lambda1_ms;
double lambda2_ms;
// parameters for the FS neurons
int FSspikecount = 0;
int SEQind_fs;
int SpikeEventCycle_FSGABA = 0;
int maxSEQdelay_FSGABA = initSEQ_FSGABA_dims[1] - 1;
int SpikeQueSize_FS = N_FS * (maxSEQdelay_FSGABA+1);
double lambda1_fs;
double lambda2_fs;
// parameters for the MS neuron spike-event generator
float p_MSSEG; // adjust for time in msec, not seconds!
int *spk_ms; spk_ms = (int *)malloc(N_MS * sizeof(int));
float U_MSSEG;
int R_MSSEG;
double gluExp_ms_AMPA = exp(-dt / ts_glu_AMPA);
double gluExp_ms_NMDA = exp(-dt / ts_glu_NMDA);
double gabaExp_ms = exp(-dt / ts_gaba);
float EMg = 1; // magnesium ion concentration in mM
double SpksExp = exp(-dt / ts_spks);
// parameters for the FS neuron spike-event generator
float p_FSSEG; // adjust for time in msec, not seconds!
int *spk_fs; spk_fs = (int *)malloc(N_FS * sizeof(int));
float U_FSSEG;
int R_FSSEG;
double gluExp_fs = exp(-dt / ts_glu_AMPA);
double gabaExp_fs = exp(-dt / ts_gaba);
// ===============================================================
// The Outputs
double *tout; // Time stamps
double *Vmsout; // example MS membrane potential
double *Vfsout; // Example FS membran potential
double *STms; // MS neuron spike times
double *STfs; // FS neuron spike times
// output states so we can continue simulations
double *Vms; // MS membrane potentials
double *Ums; // MS U values
double *Vfs; // FS membrane potentials
double *Ufs; // FS U values
double *Vfsgap; // Membrane potentials of the FS gap junctions
double *Gglu_ms_AMPA; // *** Place holder for current glutamate input conductance to the MS neurons ***
double *Gglu_ms_NMDA; // *** Place holder for current glutamate input conductance to the MS neurons ***
double *Ggaba_ms; // Current input GABA conductance to the MS neurons
double *SEQ_MSGABA; // Incomming GABAergic spike buffer for the FS neurons
double *Gglu_fs; // *** Place holder for current glutamate input conductance to the MS neurons ***
double *Ggaba_fs; // Current input GABA conductance to the MS neurons
double *SEQ_FSGABA; // Incomming GABAergic spike buffer for the FS neurons
double *RecordChan_MS_out; // output from the recording electrode
//create the output arrays
plhs[0] = mxCreateDoubleMatrix(tend, 1, mxREAL);
tout = mxGetPr(plhs[0]);
plhs[1] = mxCreateDoubleMatrix(tend, 1, mxREAL);
Vmsout = mxGetPr(plhs[1]);
plhs[2] = mxCreateDoubleMatrix(tend, 1, mxREAL);
Vfsout = mxGetPr(plhs[2]);
plhs[3] = mxCreateDoubleMatrix(MSspikebuffer, 2, mxREAL);
STms = mxGetPr(plhs[3]);
plhs[4] = mxCreateDoubleMatrix(FSspikebuffer, 2, mxREAL);
STfs = mxGetPr(plhs[4]);
plhs[5] = mxCreateNumericArray(ndim_ms,dims_ms,mxDOUBLE_CLASS,mxREAL);
Vms = mxGetPr(plhs[5]);
plhs[6] = mxCreateNumericArray(ndim_ms,dims_ms,mxDOUBLE_CLASS,mxREAL);
Ums = mxGetPr(plhs[6]);
plhs[7] = mxCreateNumericArray(ndim_fs,dims_fs,mxDOUBLE_CLASS,mxREAL);
Vfs = mxGetPr(plhs[7]);
plhs[8] = mxCreateNumericArray(ndim_fs,dims_fs,mxDOUBLE_CLASS,mxREAL);
Ufs = mxGetPr(plhs[8]);
plhs[9] = mxCreateNumericArray(ndim_fsgap,dims_fsgap,mxDOUBLE_CLASS,mxREAL);
Vfsgap = mxGetPr(plhs[9]);
plhs[10] = mxCreateNumericArray(ndim_ms,dims_ms,mxDOUBLE_CLASS,mxREAL);
Gglu_ms_AMPA = mxGetPr(plhs[10]);
plhs[11] = mxCreateNumericArray(ndim_ms,dims_ms,mxDOUBLE_CLASS,mxREAL);
Gglu_ms_NMDA = mxGetPr(plhs[11]);
plhs[12] = mxCreateNumericArray(ndim_ms,dims_ms,mxDOUBLE_CLASS,mxREAL);
Ggaba_ms = mxGetPr(plhs[12]);
plhs[13] = mxCreateNumericArray(initSEQ_MSGABA_ndim,initSEQ_MSGABA_dims,mxDOUBLE_CLASS,mxREAL);
SEQ_MSGABA = mxGetPr(plhs[13]);
plhs[14] = mxCreateNumericArray(ndim_fs,dims_fs,mxDOUBLE_CLASS,mxREAL);
Gglu_fs = mxGetPr(plhs[14]);
plhs[15] = mxCreateNumericArray(ndim_fs,dims_fs,mxDOUBLE_CLASS,mxREAL);
Ggaba_fs = mxGetPr(plhs[15]);
plhs[16] = mxCreateNumericArray(initSEQ_FSGABA_ndim,initSEQ_FSGABA_dims,mxDOUBLE_CLASS,mxREAL);
SEQ_FSGABA = mxGetPr(plhs[16]);
plhs[17] = mxCreateNumericArray(2, dims_RecordChan_MS_out,mxDOUBLE_CLASS,mxREAL);
RecordChan_MS_out = mxGetPr(plhs[17]);
// ===============================================================
// Set up easy access to the cortical state and recording channels
const int *CTX_state_dims = mxGetDimensions(prhs[58]);
#define _CTX_state(__m,__n) (CTX_state[__n*CTX_state_dims[0]+__m])
const int *RecordChan_MS_out_dims = mxGetDimensions(plhs[17]);
#define _RecordChan_MS_out(__m,__n) (RecordChan_MS_out[__n*RecordChan_MS_out_dims[0]+__m])
// ===============================================================
// init the variables
for (int i = 0; i < N_MS; i++){
Vms[i] = initVms.data[i];
Ums[i] = initUms.data[i];
}
for (int i = 0; i < N_FS; i++){
Vfs[i] = initVfs.data[i];
Ufs[i] = initUfs.data[i];
}
for (int i = 0; i < N_fsgap; i++){
Vfsgap[i] = initVfsgap.data[i];
}
float *IAMPA_ms; IAMPA_ms = (float *)malloc(N_MS * sizeof(float));
float *INMDA_ms; INMDA_ms = (float *)malloc(N_MS * sizeof(float));
float *IDA_ms; IDA_ms = (float *)malloc(N_MS * sizeof(float));
float *IGABA_ms; IGABA_ms = (float *)malloc(N_MS * sizeof(float));
float *Isyn_ms; Isyn_ms = (float *)malloc(N_MS * sizeof(float));
float tmpIDA_ms;
for (int i = 0; i < N_MS; i++){
IAMPA_ms[i] = 0;
INMDA_ms[i] = 0;
IDA_ms[i] = 0;
IGABA_ms[i] = 0;
Isyn_ms[i] = 0;
tmpIDA_ms = 0;
}
float *IAMPA_fs; IAMPA_fs = (float *)malloc(N_FS * sizeof(float));
float *IDA_fs; IDA_fs = (float *)malloc(N_FS * sizeof(float));
float *IGABA_fs; IGABA_fs = (float *)malloc(N_FS * sizeof(float));
float *Isyn_fs; Isyn_fs = (float *)malloc(N_FS * sizeof(float));
float *Igapfs; Igapfs = (float *)malloc(N_FS * sizeof(float));
for (int j = 0; j < N_FS; j++){
IAMPA_fs[j] = 0.0;
IGABA_fs[j] = 0.0;
IDA_fs[j] = 0.0;
Isyn_fs[j] = 0.0;
Igapfs[j] = 0.0;
}
int Vgap1,Vgap2;
// ===============================================================
// Active D1 and D2 receptors for the MS and FS neurons
lambda1_ms = DA; lambda2_ms = DA;
lambda1_fs = DA; lambda2_fs = DA;
// ===============================================================
// run the simulation
int SelectionCounter = 0;
for (int t = (int)tstart; t < tend; t++){
T = t * dt;
// reset the random seed
if (UNI < (0.01*dt)){
RANDSEED++;
zigset(RANDSEED);
}
// ================================================================
// Look for changes in the selection pulses
if (Pt.data[SelectionCounter] == t){
printf("Selection Pulse Detected %f \n", T);
if (Pch.data[SelectionCounter] == 1){
printf("Selection Pulse Channel %i \n", Pch.data[SelectionCounter]);
for (int i = 0; i < CHAN1_MS.M; i++){
r_MSSEG.data[CHAN1_MS.data[i]] = Phz.data[SelectionCounter];
}
for (int i = 0; i < CHAN1_FS.M; i++){
r_FSSEG.data[CHAN1_FS.data[i]] = Phz.data[SelectionCounter];
}
}
if (Pch.data[SelectionCounter] == 2){
printf("Selection Pulse Channel %i \n", Pch.data[SelectionCounter]);
for (int i = 0; i < CHAN2_MS.M; i++){
r_MSSEG.data[CHAN2_MS.data[i]] = Phz.data[SelectionCounter];
}
for (int i = 0; i < CHAN2_FS.M; i++){
r_FSSEG.data[CHAN2_FS.data[i]] = Phz.data[SelectionCounter];
}
}
SelectionCounter++;
}
// ===============================================================
// Update the MS neurons
for (int i = 0; i < N_MS; i++){
// ---------------------------------------------------------------
// Update the MS spike-event que
spk_ms[i] = 0; // reset the spike que
int T_MSSEG = 0;
p_MSSEG = r_MSSEG.data[i] * dt * 0.001;
U_MSSEG = UNI;
R_MSSEG = (U_MSSEG <= p_MSSEG); // the reference spike train
for (int ist = 0; ist < N_MSSEG.data[i]; ist++){
if (UNI <= alpha_MSSEG.data[i]){
T_MSSEG = R_MSSEG;
}
else {
T_MSSEG = (UNI <= p_MSSEG);
}
spk_ms[i] = spk_ms[i] + T_MSSEG;
}
// ---------------------------------------------------------------
//Update the MS neurons cortical input
Gglu_ms_AMPA[i] = Gglu_ms_AMPA[i] + ((Cctms_w.data[i]) * spk_ms[i] / ts_glu_AMPA);
Gglu_ms_AMPA[i] = Gglu_ms_AMPA[i] * gluExp_ms_AMPA;
Gglu_ms_NMDA[i] = Gglu_ms_NMDA[i] + (glu_ratio * Cctms_w.data[i] * spk_ms[i] / ts_glu_NMDA);
Gglu_ms_NMDA[i] = Gglu_ms_NMDA[i] * gluExp_ms_NMDA;
// ---------------------------------------------------------------
// Mg block of the MS NMDA channels
Mg_ms[i] = 1 / ( 1 + (EMg / 3.57) * exp(-Vms[i] * 0.062) );
// ---------------------------------------------------------------
// get PSPs from the other MS and FS neurons
Ggaba_ms[i] += SEQ_MSGABA[SpikeEventCycle_MSGABA*N_MS + i]; // Adds current GABAergic PSPs to the conductance bin
Ggaba_ms[i] = Ggaba_ms[i] * gabaExp_ms;
SEQ_MSGABA[SpikeEventCycle_MSGABA*N_MS + i] = 0; // reset the PSP buffer
// ---------------------------------------------------------------
// update the membrane
VMS = Vms[i]; // save the previous state
UMS = Ums[i]; // save the previous state
IGABA_ms[i] = (Ggaba_ms[i] * (Egaba - VMS));
IAMPA_ms[i] = (Gglu_ms_AMPA[i] * (Eglu - VMS)) * (1 + _MS(i,12) * lambda1_ms);
INMDA_ms[i] = Mg_ms[i] * (Gglu_ms_NMDA[i] * (Eglu - VMS)) * (1 + _MS(i,11) * lambda2_ms);
IDA_ms[i] = lambda1_ms * _MS(i,13) * (Vms[i] - _MS(i,9));
Isyn_ms[i] = IAMPA_ms[i] + INMDA_ms[i] + IGABA_ms[i] + IDA_ms[i];
//Vms[i] = VMS + dt * ( ( _MS(i,8) * (1 - _MS(i,10)*lambda2_ms) * (VMS - _MS(i,1)) * (VMS - _MS(i,2)) - Ums[i]) + Isyn_ms[i] ) / _MS(i,0);
//Ums[i] = Ums[i] + dt * _MS(i,3) * (_MS(i,4) * (VMS - _MS(i,1)) - Ums[i]);
Vms1 = VMS + hdt * ( ( _MS(i,8) * (1 - _MS(i,10)*lambda2_ms) * (VMS - _MS(i,1)) * (VMS - _MS(i,2)) - UMS) + Isyn_ms[i] ) / _MS(i,0);
Ums1 = UMS + hdt * _MS(i,3) * (_MS(i,4) * (VMS - _MS(i,1)) - UMS);
Vms[i] = VMS + dt * (( _MS(i,8) * (1 - _MS(i,10)*lambda2_ms) * (Vms1 - _MS(i,1)) * (Vms1 - _MS(i,2)) - Ums1) + Isyn_ms[i] ) / _MS(i,0);
Ums[i] = UMS + dt * _MS(i,3) * (_MS(i,4) * (Vms1 - _MS(i,1)) - Ums1);
if (Vms[i] >= _MS(i,7)){ // Check for spike events
VMS = _MS(i,7);
Vms[i] = _MS(i,5);
Ums[i] = Ums[i] + _MS(i,6);
// ---------------------------------------------------------------
// save the cell number and spike time
if (MSspikecount < MSspikebuffer){
STms[MSspikecount] = i;
STms[MSspikecount+MSspikebuffer] = T;
MSspikecount++;
}
else if (MSspikecount >= MSspikebuffer){
printf("Warning, exceeded MS spike buffer size \n");
printf("%i %i \n", MSspikecount, MSspikebuffer);
t = tend;
}
// ---------------------------------------------------------------
// Update the Spike event que for the target cells
for (int targcellind = Cmsms_b.data[i]; targcellind <= Cmsms_b.data[i+1]-1; targcellind++){ // for each of the target cells
// get the index in the spike que for the target cell, given its delay
SEQind_ms = Cmsms.data[targcellind] + ((SpikeEventCycle_MSGABA-1) + Cmsms_d.data[targcellind])*N_MS;
if (SEQind_ms >= SpikeQueSize_MS){ // if SEind is larger than the spike que, wrap around
SEQind_ms = SEQind_ms - SpikeQueSize_MS;
}
// add the spike event to the spike que for the target cell
SEQ_MSGABA[SEQind_ms] = SEQ_MSGABA[SEQind_ms] + (Cmsms_w.data[targcellind] / ts_gaba); // update the Spike event que
}
}
}
// update the FS gap junctions
for (int k = 0; k < N_fsgap; k++){
Vgap1 = Pgapfs.data[k];
Vgap2 = Pgapfs.data[k + N_fsgap];
Vfsgap[k] = Vfsgap[k] + dt * ((Vfs[Vgap1] - Vfsgap[k]) + (Vfs[Vgap2] - Vfsgap[k])) / tau_fsgap;
}
// ---------------------------------------------------------------
// for each FS neuron
// ---------------------------------------------------------------
for (int j = 0; j < N_FS; j++){
float testIgapfs = Igapfs[j];
spk_fs[j] = 0; // reset the spike que
int T_FSSEG = 0;
p_FSSEG = r_FSSEG.data[j] * dt * 0.001;
U_FSSEG = UNI;
R_FSSEG = (U_FSSEG <= p_FSSEG); // the reference spike train
for (int ist = 0; ist < N_FSSEG.data[j]; ist++){
if (UNI <= alpha_FSSEG.data[j]){
T_FSSEG = R_FSSEG;
}
else {
T_FSSEG = (UNI <= p_FSSEG);
}
spk_fs[j] = spk_fs[j] + T_FSSEG;
}
// ---------------------------------------------------------------
//Update the FS neurons cortical input
Gglu_fs[j] = Gglu_fs[j] + (Cctfs_w.data[j] * spk_fs[j] / ts_glu_AMPA);
Gglu_fs[j] = Gglu_fs[j] * gluExp_fs;
// ---------------------------------------------------------------
// get GABAergic input from the other FS neurons
Ggaba_fs[j] += SEQ_FSGABA[SpikeEventCycle_FSGABA*N_FS + j]; // Adds current GABAergic PSPs to the conductance bin
Ggaba_fs[j] = Ggaba_fs[j] * gabaExp_fs;
if (SEQ_FSGABA[SpikeEventCycle_FSGABA*N_FS + j] > 1000){
printf("Too many spikes! %i %i \n", SEQ_FSGABA[SpikeEventCycle_FSGABA*N_FS + j], t);
throw "Too many spikes";
}
SEQ_FSGABA[SpikeEventCycle_FSGABA*N_FS + j] = 0; // reset the PSP buffer
// ---------------------------------------------------------------
// calculate the current from the gap junctions
Igapfs[j] = 0.0;
for (int source = Cgapfs_b.data[j]; source < Cgapfs_b.data[j+1]; source++){
Igapfs[j] = Igapfs[j] + Cgapfs_w.data[source] * (Vfsgap[Cgapfs.data[source]] - Vfs[j]);
if (isinf(Igapfs[j])){
printf("\n \n **** Warning, Igapfs is a inf. Vfsgap %f Cgapfs %i source %i **** \n \n", Vfsgap[Cgapfs.data[source]], Cgapfs.data[source], source);
throw "Igapfs is a Inf";
}
}
// ---------------------------------------------------------------
//Update the FS neurons
VFS = Vfs[j]; // save the previous state
UFS = Ufs[j]; // save the previous state
// DA modulation of the GABA currents
IAMPA_fs[j] = (Gglu_fs[j] * (Eglu - Vfs[j]));
IGABA_fs[j] = (Ggaba_fs[j] * (Egaba - Vfs[j])) * (1 - _FS(j, 11)*lambda2_fs);
Isyn_fs[j] = IAMPA_fs[j] + IGABA_fs[j];
if (isnan(Isyn_fs[j])){
printf("Warning, Isyn_fs is a NaN %f, IAMPA = %f, IGABA = %f IDA_fs = %f\n", Isyn_fs[j], IAMPA_fs[j], IGABA_fs[j], IDA_fs[j]);
printf("Warning, Vfs is a NaN \n");
throw "Vfs is a NaN";
}
if (isnan(Igapfs[j])){
printf("Warning, Igapfs is a NaN \n");
throw "Igapfs is a NaN";
}
if (isinf(Igapfs[j])){
printf("Warning, Igapfs is a inf \n");
throw "Igapfs is a Inf";
}
// Vfs[j] = VFS + dt * (_FS(j, 3) * ((VFS - _FS(j, 1) * (1-_FS(j, 10)*lambda1_fs)) * (VFS - _FS(j, 2))) - Ufs[j] + Igapfs[j] + Isyn_fs[j] ) / _FS(j, 0);
// midpoint estimate of V
Vfs1 = VFS + hdt * (_FS(j, 3) * ((VFS - _FS(j, 1) * (1-_FS(j, 10)*lambda1_fs)) * (VFS - _FS(j, 2))) - UFS + Igapfs[j] + Isyn_fs[j] ) / _FS(j, 0);
// midpoint estimate of U
if (VFS < _FS(j, 9)){
//Ufs[j] = UFS + dt * -_FS(j, 4)*UFS;
Ufs1 = UFS + hdt * -_FS(j, 4)*UFS;
}
else if (VFS >= _FS(j, 9)){
tmp = (VFS - _FS(j, 9));
//Ufs[j] = UFS + dt * _FS(j, 4) * (_FS(j, 5) * pow((VFS - _FS(j, 9)), 3) - UFS);
Ufs1 = UFS + hdt * _FS(j, 4) * (_FS(j, 5) * tmp*tmp*tmp - UFS);
}
// second-step update of V
Vfs[j] = VFS + dt * (_FS(j, 3) * ((Vfs1 - _FS(j, 1) * (1-_FS(j, 10)*lambda1_fs)) * (Vfs1 - _FS(j, 2))) - Ufs1 + Igapfs[j] + Isyn_fs[j] ) / _FS(j, 0);
// second-step update of U
if (VFS < _FS(j, 9)){
//Ufs[j] = UFS + dt * -_FS(j, 4)*UFS;
Ufs[j] = UFS + dt * -_FS(j, 4)*Ufs1;
}
else if (VFS >= _FS(j, 9)){
tmp = (Vfs1 - _FS(j, 9));
//Ufs[j] = UFS + dt * _FS(j, 4) * (_FS(j, 5) * pow((VFS - _FS(j, 9)), 3) - UFS);
Ufs[j] = UFS + dt * _FS(j, 4) * (_FS(j, 5) * tmp*tmp*tmp - Ufs1);
}
if (isnan(Vfs[j])){
printf("Warning, Isyn_fs = %f, and Igapfs = %f \n", Isyn_fs[j], Igapfs[j]);
printf("Warning, Vfs is a NaN \n");
throw "Vfs is a NaN";
}
// ---------------------------------------------------------------
// Check for spike events
if (Vfs[j] >= _FS(j, 8)){
VFS = _FS(j, 8);
Vfs[j] = _FS(j, 6);
Ufs[j] = Ufs[j] + _FS(j, 7);
// ---------------------------------------------------------------
// save the cell number and spike time
if (FSspikecount < FSspikebuffer){
STfs[FSspikecount] = j;
STfs[FSspikecount+FSspikebuffer] = T;
FSspikecount++;
}
// check for overrun of the spike buffer
else if (FSspikecount >= FSspikebuffer){
printf("Warning, exceeded FS spike buffer size \n");
throw "exceeded FS spike buffer size";
}
// ---------------------------------------------------------------
// Update the MS spike event que
for (int targcellind = Cfsms_b.data[j]; targcellind <= Cfsms_b.data[j+1]-1; targcellind++){ // for each of the target cells
// get the index in the spike que for the target cell, given its delay
SEQind_ms = Cfsms.data[targcellind] + ((SpikeEventCycle_MSGABA-1) + Cfsms_d.data[targcellind])*N_MS;
if (SEQind_ms >= SpikeQueSize_MS){ // if SEind is larger than the spike que, wrap around
SEQind_ms = SEQind_ms - SpikeQueSize_MS;
}
// add the spike event to the spike que for the target cell
SEQ_MSGABA[SEQind_ms] = SEQ_MSGABA[SEQind_ms] + (Cfsms_w.data[targcellind] / ts_gaba); // update the Spike event que
}
// ---------------------------------------------------------------
// Update the FS spike event que
for (int targcellind = Cfsfs_b.data[j]; targcellind <= Cfsfs_b.data[j+1]-1; targcellind++){ // for each of the target cells
// get the index in the spike que for the target cell, given its delay
SEQind_fs = Cfsfs.data[targcellind] + ((SpikeEventCycle_FSGABA-1) + Cfsfs_d.data[targcellind])*N_FS;
// if SEind is larger than the spike que, wrap around
if (SEQind_fs >= SpikeQueSize_FS){
SEQind_fs = SEQind_fs - SpikeQueSize_FS;
}
if ((Cfsfs_w.data[targcellind]) > 100){
printf("Warning, FS weight too big! %i %f \n", targcellind, Cfsfs_w.data[targcellind]);
throw "FS weight too big!";
}
// add the spike event to the spike que for the target cell
SEQ_FSGABA[SEQind_fs] = SEQ_FSGABA[SEQind_fs] + (Cfsfs_w.data[targcellind] / ts_gaba);
}
}
} // for each FS neurons
// ===============================================================
// update the MS PSP cycle position
if (SpikeEventCycle_MSGABA < maxSEQdelay_MSGABA){
SpikeEventCycle_MSGABA++;
}
else {
SpikeEventCycle_MSGABA = 0;
}
// update the FS PSP cycle position
if (SpikeEventCycle_FSGABA < maxSEQdelay_FSGABA){
SpikeEventCycle_FSGABA++;
}
else {
SpikeEventCycle_FSGABA = 0;
}
// ===============================================================
// save some example neuron behaviour
tout[t] = T;
Vmsout[t] = Vms[0];
Vfsout[t] = Vfs[0];
for (int RecInd = 0; RecInd < RecordChan_MS.M; RecInd++){
_RecordChan_MS_out(t, RecInd) = Isyn_ms[RecordChan_MS.data[RecInd]];
}
_RecordChan_MS_out(t, 0) = Vms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 1) = Ums[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 2) = spk_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 3) = Gglu_ms_AMPA[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 4) = Gglu_ms_NMDA[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 5) = IAMPA_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 6) = INMDA_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 7) = Ggaba_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 8) = IGABA_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 9) = IDA_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 10) = Isyn_ms[RecordChan_MS.data[1]];
_RecordChan_MS_out(t, 11) = Vms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 12) = Ums[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 13) = spk_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 14) = Gglu_ms_AMPA[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 15) = Gglu_ms_NMDA[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 16) = IAMPA_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 17) = INMDA_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 18) = Ggaba_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 19) = IGABA_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 20) = IDA_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 21) = Isyn_ms[RecordChan_MS.data[10]];
_RecordChan_MS_out(t, 22) = Vms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 23) = Ums[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 24) = spk_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 25) = Gglu_ms_AMPA[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 26) = Gglu_ms_NMDA[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 27) = IAMPA_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 28) = INMDA_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 29) = Ggaba_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 30) = IGABA_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 31) = IDA_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 32) = Isyn_ms[RecordChan_MS.data[15]];
_RecordChan_MS_out(t, 33) = Vms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 34) = Ums[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 35) = spk_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 36) = Gglu_ms_AMPA[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 37) = Gglu_ms_NMDA[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 38) = IAMPA_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 39) = INMDA_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 40) = Ggaba_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 41) = IGABA_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 42) = IDA_ms[RecordChan_MS.data[20]];
_RecordChan_MS_out(t, 43) = Isyn_ms[RecordChan_MS.data[20]];
} // end of the simulation loop
} // end of the main simulation function
// ========================================================================
// Code to catch exceptions before they crash DCE!!!
// ========================================================================
#include <exception>
using namespace std;
void mexFunction(int nlhs, mxArray *plhs[],
int nrhs, const mxArray *prhs[]) {
try {
// dodgy code
execute(nlhs, plhs, nrhs, prhs);
}
catch(std::exception& e) {
printf(e.what());
return;
}
catch(const char* e) {
printf(e);
return;
}
catch(...) {
// report and close gracefully
printf("an unexpected exception occurred\n");
return;
}
}