{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "### FRAP geometries" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from fem_sol import frap_solver\n", "from matplotlib import rc, rcParams\n", "import fem_utils\n", "import matplotlib.pyplot as plt\n", "import numpy as np\n", "import pandas as pd\n", "import seaborn as sns\n", "fol = '/Users/hubatsch/Desktop/DropletFRAP/Latex/Figures/'\n", "sns.set_style(\"ticks\")\n", "rcParams['axes.linewidth'] = 0.75\n", "rcParams['xtick.major.width'] = 0.75\n", "rcParams['ytick.major.width'] = 0.75" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import pylab as pl\n", "params = {'legend.fontsize': 9,\n", " 'legend.handlelength': 1}\n", "pl.rcParams.update(params)\n", "\n", "def nice_fig(xla, yla, xli, yli, size, fs=12): \n", " rc('font',**{'family':'sans-serif','sans-serif':['Helvetica']})\n", " plt.gcf().set_size_inches(size[0], size[1])\n", " plt.xlabel(xla,fontsize=fs) \n", " plt.ylabel(yla,fontsize=fs)\n", " plt.xlim(xli)\n", " plt.ylim(yli)\n", " plt.tick_params(axis='both', which='major', labelsize=fs)\n", "\n", "def save_nice_fig(name):\n", " plt.savefig(name, format='pdf', dpi=300, bbox_inches='tight',\n", " transparent=True)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "me = ['Meshes/multi_drop_gauss.xml', 'Meshes/multi_drop_gauss_med.xml',\n", " 'Meshes/multi_drop_gauss_far.xml', 'Meshes/multi_drop_gauss.xml',\n", " 'Meshes/multi_drop_gauss_med.xml', 'Meshes/multi_drop_gauss_far.xml']\n", "point_lists = [[[4, 4.5, 0.5], [4, 3.5, 0.5], [3.5, 4, 0.5], [4.5, 4, 0.5]],\n", " [[4, 5, 0.5], [4, 3, 0.5], [3, 4, 0.5], [5, 4, 0.5]],\n", " [[4, 5.5, 0.5], [4, 2.5, 0.5], [2.5, 4, 0.5], [5.5, 4, 0.5]],\n", " [[4, 4.5, 0.5], [4, 3.5, 0.5], [3.5, 4, 0.5], [4.5, 4, 0.5]],\n", " [[4, 5, 0.5], [4, 3, 0.5], [3, 4, 0.5], [5, 4, 0.5]],\n", " [[4, 5.5, 0.5], [4, 2.5, 0.5], [2.5, 4, 0.5], [5.5, 4, 0.5]]]\n", "phi_tot_int = [.99, .99, .99, .9, .9, .9]\n", "phi_tot_ext = [.01, .01, .01, .1, .1, .1]\n", "G_in = [1, 1, 1, .1, .1, .1]\n", "G_out = [1, 1, 1, 0.99/0.9, 0.99/0.9, 0.99/0.9]\n", "\n", "f_i = []\n", "\n", "for p, m, p_i, p_e, G_i, G_o in zip(point_lists, me, phi_tot_int,\n", " phi_tot_ext, G_in, G_out):\n", " f = frap_solver([4, 4, 0.5], m, name='FRAP_multi_'+m[:-4]+str(G_i), point_list=p,\n", " T=50, phi_tot_int=p_i, phi_tot_ext=p_e, G_in=G_i, G_out=G_o)\n", " f.solve_frap()\n", " f_i.append(f)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "alphas = np.linspace(0,2*np.pi, 20)\n", "ns = np.c_[np.cos(alphas), np.sin(alphas), np.zeros(len(alphas))]\n", "eps = np.linspace(0, 0.23, 100)\n", "profs = []\n", "for i in range(len(f_i)):\n", "# if i>2:\n", " profs.append([])\n", " for j in range(50):\n", " values=[]\n", " fs = fem_utils.load_time_point(f_i[i].name+'t_p_'+str(j)+'.h5',\n", " f_i[i].mesh)\n", " for n in ns:\n", " values.append([fs([4, 4, 0.5]+e*n) for e in eps])\n", " profs[i].append(np.mean(np.transpose(values), 1))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "np.savetxt('t_p.csv', profs, delimiter=',')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "ft = f_i[1]\n", "meta_data = np.r_[ft.dt, ft.T, eps]\n", "np.savetxt('meta_data.csv', meta_data, delimiter=',')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "plt.plot(eps,np.transpose(profs)[:,:])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "nice_fig('t [s]', 'intensity (a.u)', [0,50], [0,1.1], [1.5,2])\n", "plt.plot([np.mean(x)/f_i[0].phi_tot_int for x in profs[0]],\n", " lw=2, label='d=0.5', ls='-')\n", "plt.plot([np.mean(x)/f_i[1].phi_tot_int for x in profs[1]],\n", " lw=2, label='d=1', ls='--')\n", "plt.plot([np.mean(x)/f_i[2].phi_tot_int for x in profs[2]],\n", " lw=2, label='d=1.5', ls=':')\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "plt.title('$\\Phi_{out}=0.01}$', size=12)\n", "plt.gca().get_yaxis().set_visible(False)\n", "save_nice_fig(fol+'Fig3/tot_recov_neighbours_bad.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "nice_fig('t [s]', 'intensity (a.u)', [0,30], [0,1.1], [1.5,2])\n", "plt.plot([np.mean(x)/f_i[3].phi_tot_int for x in profs[3]],\n", " lw=2, label='d=0.5', ls='-')\n", "plt.plot([np.mean(x)/f_i[4].phi_tot_int for x in profs[4]],\n", " lw=2, label='d=1', ls='--')\n", "plt.plot([np.mean(x)/f_i[5].phi_tot_int for x in profs[5]],\n", " lw=2, label='d=1.5', ls=':')\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "plt.title('$\\Phi_{out}=0.1}$', size=12)\n", "plt.legend(prop={'size': 9}, frameon=False)\n", "save_nice_fig(fol+'Fig3/tot_recov_neighbours_good.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "nice_fig('x [$\\mu m$]', 'intensity (a.u)', [0,0.25], [0,1.1], [3.8,2])\n", "l_sim = plt.plot(eps, np.transpose(profs[0])[:,::8]/f_i[0].phi_tot_int, '#1f77b4', lw=2.5)\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "l_fit = plt.plot(np.linspace(0, 0.23, 100), np.transpose(ml)[:,::8],\n", " ls='--', c='orange', lw=1.5)\n", "plt.legend([l_sim[0], l_fit[0]], ['Simulation', 'Fit'], prop={'size': 9}, frameon=False)\n", "save_nice_fig(fol+'Fig3/spat_recov_neighbours.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Define parameters for all simulations\n", "point_list = [[4, 4, 0.5], [4, 4, 1.5], [4, 4, 4],\n", " [4, 4, 0.5], [4, 4, 1.5], [4, 4, 4]]\n", "me = ['coverslip.xml', '1_5.xml', 'symmetric.xml',\n", " 'coverslip.xml', '1_5.xml', 'symmetric.xml']\n", "phi_tot_int = [.99, .99, .99, .9, .9, .9]\n", "phi_tot_ext = [.01, .01, .01, .1, .1, .1]\n", "G_in = [1, 1, 1, .1, .1, .1]\n", "G_out = [1, 1, 1, 0.99/0.9, 0.99/0.9, 0.99/0.9]\n", "f_cs = []\n", "\n", "# Zip all parameters, iterate\n", "for p, m, p_i, p_e, G_i, G_o in zip(point_list, me, phi_tot_int,\n", " phi_tot_ext, G_in, G_out):\n", " f_cs.append(frap_solver(p, 'Meshes/single_drop_'+m,\n", " name='FRAP_'+m[:-4]+str(G_i), T=60, phi_tot_int=p_i,\n", " phi_tot_ext=p_e, G_in=G_i, G_out=G_o))\n", " f_cs[-1].solve_frap()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "z = [0.5, 1.5, 4, 0.5, 1.5, 4]\n", "profs_cs = []\n", "for i, z_i in enumerate(z):\n", " profs_cs.append([])\n", " for j in range(50):\n", " values=[]\n", " fs = fem_utils.load_time_point(f_cs[i].name+'t_p_'+str(j)+'.h5',\n", " f_cs[i].mesh)\n", " for n in ns:\n", " values.append([fs([4, 4, z_i]+e*n) for e in eps])\n", " profs_cs[i].append(np.mean(np.transpose(values), 1))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "np.savetxt('t_p_neighbours.csv', profs_cs[0], delimiter=',')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "nice_fig('t [s]', '', [0,50], [0,1.1], [1.5,2])\n", "ls = ['-', '--', ':']\n", "for i, f in enumerate(f_cs[0:3]):\n", " plt.plot([np.mean(x)/f.phi_tot_int for x in profs_cs[i]],\n", " label='d='+str(z[i]), ls=ls[i], lw=2)\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "plt.title('$\\Phi_{out}=0.01}$', size=12)\n", "plt.gca().get_yaxis().set_visible(False)\n", "save_nice_fig(fol+'Fig3/tot_recov_cs_bad.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "nice_fig('t [s]', 'intensity (a.u)', [0,30], [0,1.1], [1.5,2])\n", "ls = ['-', '--', ':']\n", "for i, f in enumerate(f_cs[3:]):\n", " plt.plot([np.mean(x)/f.phi_tot_int for x in profs_cs[i+3]],\n", " label='h='+str(z[i]), lw=2, ls=ls[i])\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "plt.title('$\\Phi_{out}=0.1}$', size=12)\n", "plt.legend(prop={'size': 9}, frameon=False)\n", "save_nice_fig(fol+'Fig3/tot_recov_cs_good.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "ml_neigh = np.loadtxt('/Users/hubatsch/Desktop/DropletFRAP/matlab_fit_neigh.csv',\n", " delimiter=',')\n", "nice_fig('x [$\\mu m$]', 'intensity (a.u)', [0,0.25], [0,1.1], [3.8,2])\n", "l_sim = plt.plot(eps, np.transpose(profs_cs[0])[:,::8]/f_cs[0].phi_tot_int, '#1f77b4',\n", " lw=2.5, label='Simulation')\n", "plt.plot(range(0, 100), np.ones(100), linestyle='--', color='k')\n", "l_fit = plt.plot(np.linspace(0, 0.23, 100), np.transpose(ml_neigh)[:,::8],\n", " ls='--', c='orange', lw=1.5)\n", "plt.legend([l_sim[0], l_fit[0]], ['Simulation', 'Fit'], frameon=False)\n", "save_nice_fig(fol+'Fig3/spat_recov_coverslip.pdf')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Figure 1:" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Panel: comparison PGL-3 diffusivity with Louise's viscosity**" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "louise = pd.read_csv('/Users/hubatsch/Desktop/DropletFRAP/Latex/Figures/Fig1/Louise.csv')\n", "lars = pd.read_csv('/Users/hubatsch/Desktop/DropletFRAP/Latex/Figures/Fig1/Lars.csv')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "fig, ax1 = plt.subplots()\n", "ax2 = ax1.twinx()\n", "plt.sca(ax1)\n", "sns.lineplot(x=\"conc\", y=\"D\", data=lars, color=sns.color_palette()[1])\n", "sns.scatterplot(x=\"conc\", y=\"D\", data=lars, color=sns.color_palette()[1], alpha=0.7)\n", "plt.xlabel('$c_{salt}\\; [mM]$')\n", "plt.ylabel('$D_{in} \\;[\\mu m^2\\cdot s^{-1}]$', color=sns.color_palette()[1])\n", "plt.yticks([0, 0.05, 0.1], rotation=90, color = sns.color_palette()[1])\n", "plt.ylim(0, 0.1)\n", "ax1.set_zorder(1) \n", "ax1.patch.set_visible(False)\n", "plt.sca(ax2)\n", "sns.lineplot(x=\"conc\", y=\"vis\", data=louise, color=sns.color_palette()[0], label='data from ref[xxx]')\n", "nice_fig('c_{salt} [mM]', '$\\eta^{-1} \\;[Pa\\cdot s]^{-1}$', [40,190], [0,7.24], [2.3,2])\n", "plt.yticks(color = sns.color_palette()[0])\n", "plt.ylabel('$\\eta^{-1} \\;[Pa\\cdot s]^{-1}$ ', color = sns.color_palette()[0])\n", "plt.legend(frameon=False, fontsize=9)\n", "save_nice_fig(fol+'Fig1/Lars_vs_Louise.pdf')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Panel:coacervates PLYS/ATP, CMD/PLYS**m" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "coacervates = pd.read_csv('/Users/hubatsch/Desktop/DropletFRAP/Latex/Figures/Fig1/Coacervates.csv')\n", "sns.stripplot(data=coacervates, jitter=0.35, alpha=0.8,**{'marker': '.', 'size': 8, 'edgecolor': 'black', 'color': 'k'})\n", "ax = sns.barplot(data=coacervates, facecolor=(1, 1, 1, 0), edgecolor=(0.6, 0.6, 0.6), errcolor=(0.6, 0.6, 0.6), capsize=.2, ci='sd', errwidth=1.5)\n", "plt.setp(ax.lines, zorder=100)\n", "nice_fig(None, '$D_{in} \\;[\\mu m^2\\cdot s^{-1}]$', [None, None], [0,6], [2.3,2])\n", "plt.xticks([0,1], ('CMD/PLYS', 'PLYS/ATP'), rotation=20)\n", "plt.xlim(-0.7, 1.7)\n", "save_nice_fig(fol+'Fig1/Coacervates.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "ax.patches" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Panel: time course CMD**" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "CMD = np.loadtxt(fol+'/Fig1/CMD_timecourse.csv', delimiter=',')\n", "CMD_fit = np.loadtxt(fol+'/Fig1/CMD_fit_timecourse.csv', delimiter=',')\n", "l_sim = plt.plot(CMD[:, 0], CMD[:, 1:], '#1f77b4', lw=3, label='Simulation')\n", "l_fit = plt.plot(CMD_fit[:, 0], CMD_fit[:, 1:], '--', lw=2, c=sns.color_palette()[1], label='Simulation')\n", "plt.plot(range(0, 10), np.ones(10)*np.min(CMD_fit[:, 1]), linestyle='--', color='k', lw=2)\n", "nice_fig('x [$\\mu m$]', 'intensity (a.u)', [0,np.max(CMD_fit[:, 0])], [0,0.6], [2.3,2])\n", "save_nice_fig(fol+'Fig1/CMD_spat_recov.pdf')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Panel: time course PGL-3**" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "PGL = np.loadtxt(fol+'/Fig1/PGL_timecourse.csv', delimiter=',')\n", "PGL_fit = np.loadtxt(fol+'/Fig1/PGL_fit_timecourse.csv', delimiter=',')\n", "l_sim = plt.plot(PGL[:, 0], PGL[:, 1:], '#1f77b4', lw=3, label='Simulation')\n", "l_fit = plt.plot(PGL_fit[:, 0], PGL_fit[:, 1:], '--', lw=2, c=sns.color_palette()[1], label='Simulation')\n", "plt.plot(range(0, 10), np.ones(10)*np.min(PGL_fit[:, 1]), linestyle='--', color='k', lw=2)\n", "nice_fig('x [$\\mu m$]', 'intensity (a.u)', [0,np.max(PGL_fit[:, 0])], [0,None], [2.3,2])\n", "save_nice_fig(fol+'Fig1/PGL_spat_recov.pdf')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Panel: time course total intensity**" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "PGL = np.loadtxt(fol+'/Fig1/PGL_tot.csv', delimiter=',')\n", "ATP = np.loadtxt(fol+'/Fig1/ATP_tot.csv', delimiter=',')\n", "CMD = np.loadtxt(fol+'/Fig1/CMD_tot.csv', delimiter=',')\n", "# fig, ax1 = plt.subplots()\n", "# ax2 = ax1.twiny()\n", "# plt.sca(ax1)\n", "nice_fig('$t/T_{max}$', 'intensity (a.u)', [0,200], [0,0.62], [2.3,2])\n", "# plt.sca(ax2)\n", "# ax2.tick_params(axis=\"x\",direction=\"in\")\n", "plt.plot(PGL[::10, 0]/np.max(PGL[1:-1:2, 0]), PGL[::10,1], label='PGL-3', c='#CC406E', markersize=3, alpha=0.7, lw=2)\n", "plt.plot(ATP[::1, 0]/np.max(ATP[:, 0]), ATP[::1,1], label='PLYS/ATP', c='#FF508A', markersize=3, alpha=0.7, lw=2)\n", "plt.plot(CMD[::5, 0]/np.max(CMD[:, 0]), CMD[::5,1], label='CMD/PLYS', c='#7F2845', markersize=3, alpha=0.7, lw=2)\n", "plt.legend(frameon=False, fontsize=9)\n", "plt.xlim(0, 1)\n", "save_nice_fig(fol+'Fig1/tot_recov.pdf')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.8.2" } }, "nbformat": 4, "nbformat_minor": 4 }