コード例 #1
0
    def loadTSegmentTree(self):
        '''
        Load point neuron model
        '''
        self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX,
                                                 'Ttree_segments.swc'))
        # self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX, 'L23PyrBranco.swc'))
        # capacitance and axial resistance
        self.tree.setPhysiology(0.8, 100. / 1e6)
        # ion channels
        k_chan = channelcollection.Kv3_1()

        g_k = {1: 0.766 * 1e6}
        g_k.update({n.index: 0.034*1e6 / self.tree.pathLength((1,.5), (n.index,.5)) \
                    for n in self.tree if n.index != 1})

        self.tree.addCurrent(k_chan, g_k, -85.)
        na_chan = channelcollection.Na_Ta()
        self.tree.addCurrent(na_chan, 1.71 * 1e6, 50., node_arg=[self.tree[1]])
        # fit leak current
        self.tree.fitLeakCurrent(-75., 10.)
        # set equilibirum potententials
        self.tree.setEEq(-75.)
        # set computational tree
        self.tree.setCompTree()
コード例 #2
0
    def loadBallAndStick(self):
        '''
        Load the ball and stick model

        1--4
        '''
        self.tree = PhysTree(file_n='test_morphologies/ball_and_stick.swc')
        self.tree.setPhysiology(0.8, 100. / 1e6)
        self.tree.setLeakCurrent(100., -75.)
        self.tree.setCompTree()
コード例 #3
0
    def loadBallAndStick(self):
        '''
        Load the ball and stick model

        1--4
        '''
        self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX, 'ball_and_stick.swc'))
        self.tree.setPhysiology(0.8, 100./1e6)
        self.tree.setLeakCurrent(100., -75.)
        self.tree.setCompTree()
コード例 #4
0
    def loadTree(self, reinitialize=0, segments=False):
        """
        Load the T-tree morphology in memory

          6--5--4--7--8
                |
                |
                1
        """
        if not hasattr(self, 'tree') or reinitialize:
            print('>>> loading T-tree <<<')
            fname = 'Ttree_segments.swc' if segments else 'Ttree.swc'
            self.tree = PhysTree('test_morphologies/' + fname, types=[1, 3, 4])
コード例 #5
0
ファイル: test_phystree.py プロジェクト: sunzhenyang2018/NEAT
    def loadTree(self, reinitialize=0, segments=False):
        """
        Load the T-tree morphology in memory

          6--5--4--7--8
                |
                |
                1
        """
        if not hasattr(self, 'tree') or reinitialize:
            fname = 'Ttree_segments.swc' if segments else 'Ttree.swc'
            self.tree = PhysTree(os.path.join(MORPHOLOGIES_PATH_PREFIX, fname),
                                 types=[1, 3, 4])
コード例 #6
0
    def loadTree(self, reinitialize=0):
        '''
        Load the T-tree morphology in memory

          6--5--4--7--8
                |
                |
                1
        '''
        if not hasattr(self, 'tree') or reinitialize:
            print '>>> loading T-tree <<<'
            fname = 'test_morphologies/Ttree.swc'
            self.tree = PhysTree(fname, types=[1, 3, 4])
コード例 #7
0
    def loadTTree(self):
        '''
        Load the T-tree model

          6--5--4--7--8
                |
                |
                1
        '''
        fname = os.path.join(MORPHOLOGIES_PATH_PREFIX, 'Tsovtree.swc')
        self.tree = PhysTree(fname, types=[1, 3, 4])
        self.tree.setPhysiology(0.8, 100. / 1e6)
        self.tree.fitLeakCurrent(-75., 10.)
        self.tree.setCompTree()
コード例 #8
0
    def loadTTree(self):
        '''
        Load the T-tree model

          6--5--4--7--8
                |
                |
                1
        '''
        print('>>> loading T-tree <<<')
        fname = 'test_morphologies/Tsovtree.swc'
        self.tree = PhysTree(fname, types=[1, 3, 4])
        self.tree.setPhysiology(0.8, 100. / 1e6)
        self.tree.fitLeakCurrent(-75., 10.)
        self.tree.setCompTree()
コード例 #9
0
 def loadBall(self):
     '''
     Load point neuron model
     '''
     self.tree = PhysTree(file_n='test_morphologies/ball.swc')
     # capacitance and axial resistance
     self.tree.setPhysiology(0.8, 100. / 1e6)
     # ion channels
     k_chan = channelcollection.Kv3_1()
     self.tree.addCurrent(k_chan, 0.766 * 1e6, -85.)
     na_chan = channelcollection.Na_Ta()
     self.tree.addCurrent(na_chan, 1.71 * 1e6, 50.)
     # fit leak current
     self.tree.fitLeakCurrent(-75., 10.)
     # set equilibirum potententials
     self.tree.setEEq(-75.)
     # set computational tree
     self.tree.setCompTree()
コード例 #10
0
class TestPhysTree():
    def loadTree(self, reinitialize=0):
        '''
        Load the T-tree morphology in memory

          6--5--4--7--8
                |
                |
                1
        '''
        if not hasattr(self, 'tree') or reinitialize:
            print '>>> loading T-tree <<<'
            fname = 'test_morphologies/Ttree.swc'
            self.tree = PhysTree(fname, types=[1, 3, 4])

    def testLeakDistr(self):
        self.loadTree(reinitialize=1)
        with pytest.raises(AssertionError):
            self.tree.fitLeakCurrent(e_eq_target=-75., tau_m_target=-10.)
        # test simple distribution
        self.tree.fitLeakCurrent(e_eq_target=-75., tau_m_target=10.)
        for node in self.tree:
            assert np.abs(node.c_m - 1.0) < 1e-9
            assert np.abs(node.currents['L'][0] - 1. / (10. * 1e-3)) < 1e-9
            assert np.abs(node.e_eq + 75.) < 1e-9
        # create complex distribution
        tau_distr = lambda x: x + 100.
        for node in self.tree:
            d2s = self.tree.pathLength({
                'node': node.index,
                'x': 1.
            }, (1., 0.5))
            node.fitLeakCurrent(e_eq_target=-75., tau_m_target=tau_distr(d2s))
            assert np.abs(node.c_m - 1.0) < 1e-9
            assert np.abs(node.currents['L'][0] - 1. / (tau_distr(d2s)*1e-3)) < \
                   1e-9
            assert np.abs(node.e_eq + 75.) < 1e-9
コード例 #11
0
ファイル: L23_pyramid.py プロジェクト: unibe-cns/NEAT
def getL23PyramidPas():
    """
    Return a passive model of the L2/3 pyramidal cell
    """
    cm = 1.  # uF / cm^2
    rm = 10000.  # Ohm * cm^2
    ri = 150.  # Ohm * cm
    el = -75.  # mV

    phys_tree = PhysTree('models/morphologies/L23PyrBranco.swc',
                         types=[1, 2, 3, 4])

    # set specific membrane capacitance and axial resistance
    phys_tree.setPhysiology(
        cm,  # Cm [uF/cm^2]
        ri / 1e6,  # Ra[MOhm*cm]
    )

    # passive membrane conductance
    phys_tree.setLeakCurrent(1e6 / 10000., el)
    phys_tree.setLeakCurrent(0.02 * 1e6, el, node_arg='axonal')

    return phys_tree
コード例 #12
0
class TestPhysTree():
    def loadTree(self, reinitialize=0, segments=False):
        """
        Load the T-tree morphology in memory

          6--5--4--7--8
                |
                |
                1
        """
        if not hasattr(self, 'tree') or reinitialize:
            print('>>> loading T-tree <<<')
            fname = 'Ttree_segments.swc' if segments else 'Ttree.swc'
            self.tree = PhysTree('test_morphologies/' + fname, types=[1, 3, 4])

    def testLeakDistr(self):
        self.loadTree(reinitialize=1)
        with pytest.raises(AssertionError):
            self.tree.fitLeakCurrent(-75., -10.)
        # test simple distribution
        self.tree.fitLeakCurrent(-75., 10.)
        for node in self.tree:
            assert np.abs(node.c_m - 1.0) < 1e-9
            assert np.abs(node.currents['L'][0] - 1. / (10. * 1e-3)) < 1e-9
            assert np.abs(node.e_eq + 75.) < 1e-9
        # create complex distribution
        tau_distr = lambda x: x + 100.
        for node in self.tree:
            d2s = self.tree.pathLength({
                'node': node.index,
                'x': 1.
            }, (1., 0.5))
            node.fitLeakCurrent(self.tree.channel_storage,
                                e_eq_target=-75.,
                                tau_m_target=tau_distr(d2s))
            assert np.abs(node.c_m - 1.0) < 1e-9
            assert np.abs(node.currents['L'][0] - 1. / (tau_distr(d2s)*1e-3)) < \
                   1e-9
            assert np.abs(node.e_eq + 75.) < 1e-9

    def testPhysiologySetting(self):
        self.loadTree(reinitialize=1)
        d2s = {1: 0., 4: 50., 5: 125., 6: 175., 7: 125., 8: 175.}
        # passive parameters as float
        c_m = 1.
        r_a = 100. * 1e-6
        self.tree.setPhysiology(c_m, r_a)
        for node in self.tree:
            assert np.abs(node.c_m - c_m) < 1e-10
            assert np.abs(node.r_a - r_a) < 1e-10
        # passive parameters as function
        c_m = lambda x: .5 * x + 1.
        r_a = lambda x: np.exp(0.01 * x) * 100 * 1e-6
        self.tree.setPhysiology(c_m, r_a)
        for node in self.tree:
            assert np.abs(node.c_m - c_m(d2s[node.index])) < 1e-10
            assert np.abs(node.r_a - r_a(d2s[node.index])) < 1e-10
        # passive parameters as incomplete dict
        r_a = 100. * 1e-6
        c_m = {1: 1., 4: 1.2}
        with pytest.raises(KeyError):
            self.tree.setPhysiology(c_m, r_a)
        # passive parameters as complete dict
        c_m.update({5: 1.1, 6: 0.9, 7: 0.8, 8: 1.})
        self.tree.setPhysiology(c_m, r_a)
        for node in self.tree:
            assert np.abs(node.c_m - c_m[node.index]) < 1e-10

        # equilibrium potential as float
        e_eq = -75.
        self.tree.setEEq(e_eq)
        for node in self.tree:
            assert np.abs(node.e_eq - e_eq) < 1e-10
        # equilibrium potential as dict
        e_eq = {1: -75., 4: -74., 5: -73., 6: -72., 7: -71., 8: -70.}
        self.tree.setEEq(e_eq)
        for node in self.tree:
            assert np.abs(node.e_eq - e_eq[node.index]) < 1e-10
        # equilibrium potential as function
        e_eq = lambda x: -70. + 0.1 * x
        self.tree.setEEq(e_eq)
        for node in self.tree:
            assert np.abs(node.e_eq - e_eq(d2s[node.index])) < 1e-10
        # as wrong type
        with pytest.raises(TypeError):
            self.tree.setEEq([])
            self.tree.setPhysiology([], [])

        # leak as float
        g_l, e_l = 100., -75.
        self.tree.setLeakCurrent(g_l, e_l)
        for node in self.tree:
            g, e = node.currents['L']
            assert np.abs(g - g_l) < 1e-10
            assert np.abs(e - e_l) < 1e-10
        # equilibrium potential as dict
        g_l = {1: 101., 4: 103., 5: 105., 6: 107., 7: 108., 8: 109.}
        e_l = {1: -75., 4: -74., 5: -73., 6: -72., 7: -71., 8: -70.}
        self.tree.setLeakCurrent(g_l, e_l)
        for node in self.tree:
            g, e = node.currents['L']
            assert np.abs(g - g_l[node.index]) < 1e-10
            assert np.abs(e - e_l[node.index]) < 1e-10
        # equilibrium potential as function
        g_l = lambda x: 100. + 0.05 * x
        e_l = lambda x: -70. + 0.05 * x
        self.tree.setLeakCurrent(g_l, e_l)
        for node in self.tree:
            g, e = node.currents['L']
            assert np.abs(g - g_l(d2s[node.index])) < 1e-10
            assert np.abs(e - e_l(d2s[node.index])) < 1e-10
        # as wrong type
        with pytest.raises(TypeError):
            self.tree.setLeakCurrent([])

        # gmax as potential as float
        e_rev = 100.
        g_max = 100.
        channel = channelcollection.TestChannel2()
        self.tree.addCurrent(channel, g_max, e_rev)
        for node in self.tree:
            g_m = node.currents['TestChannel2'][0]
            assert np.abs(g_m - g_max) < 1e-10
        # equilibrium potential as dict
        g_max = {1: 101., 4: 103., 5: 104., 6: 106., 7: 107., 8: 110.}
        self.tree.addCurrent(channel, g_max, e_rev)
        for node in self.tree:
            g_m = node.currents['TestChannel2'][0]
            assert np.abs(g_m - g_max[node.index]) < 1e-10
        # equilibrium potential as function
        g_max = lambda x: 100. + 0.005 * x**2
        self.tree.addCurrent(channel, g_max, e_rev)
        for node in self.tree:
            g_m = node.currents['TestChannel2'][0]
            assert np.abs(g_m - g_max(d2s[node.index])) < 1e-10
        # test is channel is stored
        assert isinstance(
            self.tree.channel_storage[channel.__class__.__name__],
            channelcollection.TestChannel2)
        # check if error is thrown if an ionchannel is not give
        with pytest.raises(IOError):
            self.tree.addCurrent('TestChannel2', g_max, e_rev)

    def testMembraneFunctions(self):
        self.loadTree(reinitialize=1)
        self.tree.setPhysiology(1., 100 * 1e-6)
        # passive parameters
        c_m = 1.
        r_a = 100. * 1e-6
        e_eq = -75.
        self.tree.setPhysiology(c_m, r_a)
        self.tree.setEEq(e_eq)
        # channel
        p_open = .9 * .3**3 * .5**2 + .1 * .4**2 * .6**1  # TestChannel2
        g_chan, e_chan = 100., 100.
        channel = channelcollection.TestChannel2()
        self.tree.addCurrent(channel, g_chan, e_chan)
        # fit the leak current
        self.tree.fitLeakCurrent(-30., 10.)

        # test if fit was correct
        for node in self.tree:
            tau_mem = c_m / (node.currents['L'][0] + g_chan * p_open) * 1e3
            assert np.abs(tau_mem - 10.) < 1e-10
            e_eq = (node.currents['L'][0]*node.currents['L'][1] + \
                    g_chan*p_open*e_chan) / (node.currents['L'][0] + g_chan*p_open)
            assert np.abs(e_eq - (-30.)) < 1e-10

        # test if warning is raised for impossible to reach time scale
        with pytest.warns(UserWarning):
            tree = copy.deepcopy(self.tree)
            tree.fitLeakCurrent(-30., 100000.)

        # total membrane conductance
        g_pas = self.tree[1].currents['L'][0] + g_chan * p_open
        # make passive membrane
        tree = copy.deepcopy(self.tree)
        tree.asPassiveMembrane()
        # test if fit was correct
        for node in tree:
            assert np.abs(node.currents['L'][0] - g_pas) < 1e-10

    def testCompTree(self):
        self.loadTree(reinitialize=1, segments=True)

        # capacitance axial resistance constant
        c_m = 1.
        r_a = 100. * 1e-6
        self.tree.setPhysiology(c_m, r_a)
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 8, 10, 12]
        # capacitance and axial resistance change
        c_m = lambda x: 1. if x < 200. else 1.6
        r_a = lambda x: 1. if x < 300. else 1.6
        self.tree.setPhysiology(c_m, r_a)
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 5, 6, 8, 10, 12]
        # leak current changes
        g_l = lambda x: 100. if x < 400. else 160.
        self.tree.setLeakCurrent(g_l, -75.)
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 5, 6, 7, 8, 10, 12]
        # leak current & reversal change
        g_l = 100.
        e_l = {ind: -75. for ind in [1, 4, 5, 6, 7, 8, 11, 12]}
        e_l.update({ind: -55. for ind in [9, 10]})
        self.tree.setLeakCurrent(g_l, e_l)
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 5, 6, 8, 10, 12]
        # leak current & reversal change
        g_l = 100.
        e_l = {ind: -75. for ind in [1, 4, 5, 6, 7, 8, 10, 11, 12]}
        e_l.update({9: -55.})
        self.tree.setLeakCurrent(g_l, e_l)
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 5, 6, 8, 9, 10, 12]
        # shunt
        self.tree.treetype = 'original'
        self.tree[7].g_shunt = 1.
        self.tree.setCompTree()
        self.tree.treetype = 'computational'
        assert [n.index for n in self.tree] == [1, 5, 6, 7, 8, 9, 10, 12]
コード例 #13
0
class TestCompartmentFitter():
    def loadTTree(self):
        '''
        Load the T-tree model

          6--5--4--7--8
                |
                |
                1
        '''
        fname = os.path.join(MORPHOLOGIES_PATH_PREFIX, 'Tsovtree.swc')
        self.tree = PhysTree(fname, types=[1, 3, 4])
        self.tree.setPhysiology(0.8, 100. / 1e6)
        self.tree.fitLeakCurrent(-75., 10.)
        self.tree.setCompTree()

    def loadBallAndStick(self):
        '''
        Load the ball and stick model

        1--4
        '''
        self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX,
                                                 'ball_and_stick.swc'))
        self.tree.setPhysiology(0.8, 100. / 1e6)
        self.tree.setLeakCurrent(100., -75.)
        self.tree.setCompTree()

    def loadBall(self):
        '''
        Load point neuron model
        '''
        self.tree = PhysTree(
            file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX, 'ball.swc'))
        # capacitance and axial resistance
        self.tree.setPhysiology(0.8, 100. / 1e6)
        # ion channels
        k_chan = channelcollection.Kv3_1()
        self.tree.addCurrent(k_chan, 0.766 * 1e6, -85.)
        na_chan = channelcollection.Na_Ta()
        self.tree.addCurrent(na_chan, 1.71 * 1e6, 50.)
        # fit leak current
        self.tree.fitLeakCurrent(-75., 10.)
        # set equilibirum potententials
        self.tree.setEEq(-75.)
        # set computational tree
        self.tree.setCompTree()

    def loadTSegmentTree(self):
        '''
        Load point neuron model
        '''
        self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX,
                                                 'Ttree_segments.swc'))
        # self.tree = PhysTree(file_n=os.path.join(MORPHOLOGIES_PATH_PREFIX, 'L23PyrBranco.swc'))
        # capacitance and axial resistance
        self.tree.setPhysiology(0.8, 100. / 1e6)
        # ion channels
        k_chan = channelcollection.Kv3_1()

        g_k = {1: 0.766 * 1e6}
        g_k.update({n.index: 0.034*1e6 / self.tree.pathLength((1,.5), (n.index,.5)) \
                    for n in self.tree if n.index != 1})

        self.tree.addCurrent(k_chan, g_k, -85.)
        na_chan = channelcollection.Na_Ta()
        self.tree.addCurrent(na_chan, 1.71 * 1e6, 50., node_arg=[self.tree[1]])
        # fit leak current
        self.tree.fitLeakCurrent(-75., 10.)
        # set equilibirum potententials
        self.tree.setEEq(-75.)
        # set computational tree
        self.tree.setCompTree()

    def testTreeStructure(self):
        self.loadTTree()
        cm = CompartmentFitter(self.tree)
        # set of locations
        fit_locs1 = [(1, .5), (4, .5), (5, .5)]  # no bifurcations
        fit_locs2 = [(1, .5), (4, .5), (5, .5),
                     (8, .5)]  # w bifurcation, should be added
        fit_locs3 = [(1, .5), (4, 1.), (5, .5),
                     (8, .5)]  # w bifurcation, already added

        # test fit_locs1, no bifurcation are added
        # input paradigm 1
        cm.setCTree(fit_locs1, extend_w_bifurc=True)

        # fl1_a = cm.tree.getLocs('fit locs')
        # with pytest.warns(UserWarning):
        #     cm.setCTree(fit_locs1, extend_w_bifurc=False)
        #     fl1_b = cm.tree.getLocs('fit locs')
        # assert len(fl1_a) == len(fl1_b)
        # for fla, flb in zip(fl1_a, fl1_b): assert fla == flb
        # # input paradigm 2
        # cm.tree.storeLocs(fit_locs1, 'fl1')
        # cm.setCTree('fl1', extend_w_bifurc=True)
        # fl1_a = cm.tree.getLocs('fit locs')
        # assert len(fl1_a) == len(fl1_b)
        # for fla, flb in zip(fl1_a, fl1_b): assert fla == flb
        # # test tree structure
        # assert len(cm.ctree) == 3
        # for cn in cm.ctree: assert len(cn.child_nodes) <= 1

        # # test fit_locs2, a bifurcation should be added
        # with pytest.warns(UserWarning):
        #     cm.setCTree(fit_locs2, extend_w_bifurc=False)
        # fl2_b = cm.tree.getLocs('fit locs')
        # cm.setCTree(fit_locs2, extend_w_bifurc=True)
        # fl2_a = cm.tree.getLocs('fit locs')
        # assert len(fl2_a) == len(fl2_b) + 1
        # for fla, flb in zip(fl2_a, fl2_b): assert fla == flb
        # assert fl2_a[-1] == (4,1.)
        # # test tree structure
        # assert len(cm.ctree) == 5
        # for cn in cm.ctree:
        #     assert len(cn.child_nodes) <= 1 if cn.loc_ind != 4 else \
        #            len(cn.child_nodes) == 2

        # # test fit_locs2, no bifurcation should be added as it is already present
        # cm.setCTree(fit_locs3, extend_w_bifurc=True)
        # fl3 = cm.tree.getLocs('fit locs')
        # for fl_, fl3 in zip(fit_locs3, fl3): assert fl_ == fl3
        # # test tree structure
        # assert len(cm.ctree) == 4
        # for cn in cm.ctree:
        #     assert len(cn.child_nodes) <= 1 if cn.loc_ind != 1 else \
        #            len(cn.child_nodes) == 2

    def _checkChannels(self, tree, channel_names):
        assert isinstance(tree, compartmentfitter.FitTreeGF)
        assert set(tree.channel_storage.keys()) == set(channel_names)
        for node in tree:
            assert set(node.currents.keys()) == set(channel_names + ['L'])

    def testCreateTreeGF(self):
        self.loadBall()
        cm = CompartmentFitter(self.tree)

        # create tree with only 'L'
        tree_pas = cm.createTreeGF()
        self._checkChannels(tree_pas, [])
        # create tree with only 'Na_Ta'
        tree_na = cm.createTreeGF(['Na_Ta'])
        self._checkChannels(tree_na, ['Na_Ta'])
        # create tree with only 'Kv3_1'
        tree_k = cm.createTreeGF(['Kv3_1'])
        self._checkChannels(tree_k, ['Kv3_1'])
        # create tree with all channels
        tree_all = cm.createTreeGF(['Na_Ta', 'Kv3_1'])
        self._checkChannels(tree_all, ['Na_Ta', 'Kv3_1'])

    def reduceExplicit(self):
        self.loadBall()

        freqs = np.array([0.])
        locs = [(1, 0.5)]
        e_eqs = [-75., -55., -35., -15.]
        # create compartment tree
        ctree = self.tree.createCompartmentTree(locs)
        ctree.addCurrent(channelcollection.Na_Ta(), 50.)
        ctree.addCurrent(channelcollection.Kv3_1(), -85.)

        # create tree with only leak
        greens_tree_pas = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_pas[1].currents = {'L': greens_tree_pas[1].currents['L']}
        greens_tree_pas.setCompTree()
        greens_tree_pas.setImpedance(freqs)
        # compute the passive impedance matrix
        z_mat_pas = greens_tree_pas.calcImpedanceMatrix(locs)[0]

        # create tree with only potassium
        greens_tree_k = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_k[1].currents = {key: val for key, val in greens_tree_k[1].currents.items() \
                                               if key != 'Na_Ta'}
        # compute potassium impedance matrices
        z_mats_k = []
        for e_eq in e_eqs:
            greens_tree_k.setEEq(e_eq)
            greens_tree_k.setCompTree()
            greens_tree_k.setImpedance(freqs)
            z_mats_k.append(greens_tree_k.calcImpedanceMatrix(locs))

        # create tree with only sodium
        greens_tree_na = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_na[1].currents = {key: val for key, val in greens_tree_na[1].currents.items() \
                                               if key != 'Kv3_1'}
        # create state variable expansion points
        svs = []
        e_eqs_ = []
        na_chan = greens_tree_na.channel_storage['Na_Ta']
        for e_eq1 in e_eqs:
            sv1 = na_chan.computeVarinf(e_eq1)
            for e_eq2 in e_eqs:
                e_eqs_.append(e_eq2)
                sv2 = na_chan.computeVarinf(e_eq2)
                svs.append({'m': sv2['m'], 'h': sv1['h']})

        # compute sodium impedance matrices
        z_mats_na = []
        for sv, eh in zip(svs, e_eqs_):
            greens_tree_na.setEEq(eh)
            greens_tree_na[1].setExpansionPoint('Na_Ta', sv)
            greens_tree_na.setCompTree()
            greens_tree_na.setImpedance(freqs)
            z_mats_na.append(greens_tree_na.calcImpedanceMatrix(locs))

        # passive fit
        ctree.computeGMC(z_mat_pas)

        # potassium channel fit matrices
        fit_mats_k = []
        for z_mat_k, e_eq in zip(z_mats_k, e_eqs):
            mf, vt = ctree.computeGSingleChanFromImpedance(
                'Kv3_1',
                z_mat_k,
                e_eq,
                freqs,
                other_channel_names=['L'],
                action='return')
            fit_mats_k.append([mf, vt])

        # sodium channel fit matrices
        fit_mats_na = []
        for z_mat_na, e_eq, sv in zip(z_mats_na, e_eqs_, svs):
            mf, vt = ctree.computeGSingleChanFromImpedance(
                'Na_Ta',
                z_mat_na,
                e_eq,
                freqs,
                sv=sv,
                other_channel_names=['L'],
                action='return')
            fit_mats_na.append([mf, vt])

        return fit_mats_na, fit_mats_k

    def testChannelFitMats(self):
        self.loadBall()
        cm = CompartmentFitter(self.tree)
        cm.setCTree([(1, .5)])
        # check if reversals are correct
        for key in set(cm.ctree[0].currents) - {'L'}:
            assert np.abs(cm.ctree[0].currents[key][1] - \
                          self.tree[1].currents[key][1]) < 1e-10

        # fit the passive model
        cm.fitPassive(use_all_channels=False)

        fit_mats_cm_na = cm.evalChannel('Na_Ta', parallel=False)
        fit_mats_cm_k = cm.evalChannel('Kv3_1', parallel=False)
        fit_mats_control_na, fit_mats_control_k = self.reduceExplicit()
        # test whether potassium fit matrices agree
        for fm_cm, fm_control in zip(fit_mats_cm_k, fit_mats_control_k):
            assert np.allclose(np.sum(fm_cm[0]),
                               fm_control[0][0, 0])  # feature matrices
            assert np.allclose(fm_cm[1], fm_control[1])  # target vectors
        # test whether sodium fit matrices agree
        for fm_cm, fm_control in zip(fit_mats_cm_na[4:], fit_mats_control_na):
            assert np.allclose(np.sum(fm_cm[0]),
                               fm_control[0][0, 0])  # feature matrices
            assert np.allclose(fm_cm[1], fm_control[1])  # target vectors

    def _checkPasCondProps(self, ctree1, ctree2):
        assert len(ctree1) == len(ctree2)
        for n1, n2 in zip(ctree1, ctree2):
            assert np.allclose(n1.currents['L'][0], n2.currents['L'][0])
            assert np.allclose(n1.g_c, n2.g_c)

    def _checkPasCaProps(self, ctree1, ctree2):
        assert len(ctree1) == len(ctree2)
        for n1, n2 in zip(ctree1, ctree2):
            assert np.allclose(n1.ca, n2.ca)

    def _checkAllCurrProps(self, ctree1, ctree2):
        assert len(ctree1) == len(ctree2)
        assert ctree1.channel_storage.keys() == ctree2.channel_storage.keys()
        for n1, n2 in zip(ctree1, ctree2):
            assert np.allclose(n1.g_c, n2.g_c)
            for key in n1.currents:
                assert np.allclose(n1.currents[key][0], n2.currents[key][0])
                assert np.allclose(n1.currents[key][1], n2.currents[key][1])

    def _checkPhysTrees(self, tree1, tree2):
        assert len(tree1) == len(tree2)
        assert tree1.channel_storage.keys() == tree2.channel_storage.keys()
        for n1, n2 in zip(tree1, tree2):
            assert np.allclose(n1.r_a, n2.r_a)
            assert np.allclose(n1.c_m, n2.c_m)
            for key in n1.currents:
                assert np.allclose(n1.currents[key][0], n2.currents[key][0])
                assert np.allclose(n1.currents[key][1], n2.currents[key][1])

    def _checkEL(self, ctree, e_l):
        for n in ctree:
            assert np.allclose(n.currents['L'][1], e_l)

    def testPassiveFit(self):
        self.loadTTree()
        fit_locs = [(1, .5), (4, 1.), (5, .5), (8, .5)]

        # fit a tree directly from CompartmentTree
        greens_tree = self.tree.__copy__(new_tree=GreensTree())
        greens_tree.setCompTree()
        freqs = np.array([0.])
        greens_tree.setImpedance(freqs)
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs)[0].real
        ctree = greens_tree.createCompartmentTree(fit_locs)
        ctree.computeGMC(z_mat)
        sov_tree = self.tree.__copy__(new_tree=SOVTree())
        sov_tree.calcSOVEquations()
        alphas, phimat = sov_tree.getImportantModes(locarg=fit_locs)
        ctree.computeC(-alphas[0:1].real * 1e3, phimat[0:1, :].real)

        # fit a tree with compartment fitter
        cm = CompartmentFitter(self.tree)
        cm.setCTree(fit_locs)
        cm.fitPassive()
        cm.fitCapacitance()
        cm.fitEEq()

        # check whether both trees are the same
        self._checkPasCondProps(ctree, cm.ctree)
        self._checkPasCaProps(ctree, cm.ctree)
        self._checkEL(cm.ctree, -75.)

        # test whether all channels are used correctly for passive fit
        self.loadBall()
        fit_locs = [(1, .5)]
        # fit ball model with only leak
        greens_tree = self.tree.__copy__(new_tree=GreensTree())
        greens_tree.channel_storage = {}
        for n in greens_tree:
            n.currents = {'L': n.currents['L']}
        greens_tree.setCompTree()
        freqs = np.array([0.])
        greens_tree.setImpedance(freqs)
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs)[0].real
        ctree_leak = greens_tree.createCompartmentTree(fit_locs)
        ctree_leak.computeGMC(z_mat)
        sov_tree = greens_tree.__copy__(new_tree=SOVTree())
        sov_tree.calcSOVEquations()
        alphas, phimat = sov_tree.getImportantModes(locarg=fit_locs)
        ctree_leak.computeC(-alphas[0:1].real * 1e3, phimat[0:1, :].real)
        # make ball model with leak based on all channels
        tree = self.tree.__copy__()
        tree.asPassiveMembrane()
        greens_tree = tree.__copy__(new_tree=GreensTree())
        greens_tree.setCompTree()
        freqs = np.array([0.])
        greens_tree.setImpedance(freqs)
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs)[0].real
        ctree_all = greens_tree.createCompartmentTree(fit_locs)
        ctree_all.computeGMC(z_mat)
        sov_tree = tree.__copy__(new_tree=SOVTree())
        sov_tree.calcSOVEquations()
        alphas, phimat = sov_tree.getImportantModes(locarg=fit_locs)
        ctree_all.computeC(-alphas[0:1].real * 1e3, phimat[0:1, :].real)

        # new compartment fitter
        cm = CompartmentFitter(self.tree)
        cm.setCTree(fit_locs)
        # test fitting
        cm.fitPassive(use_all_channels=False)
        cm.fitCapacitance()
        cm.fitEEq()
        self._checkPasCondProps(ctree_leak, cm.ctree)
        self._checkPasCaProps(ctree_leak, cm.ctree)
        with pytest.raises(AssertionError):
            self._checkEL(cm.ctree, self.tree[1].currents['L'][1])
        cm.fitPassive(use_all_channels=True)
        cm.fitCapacitance()
        cm.fitEEq()
        self._checkPasCondProps(ctree_all, cm.ctree)
        self._checkPasCaProps(ctree_all, cm.ctree)
        self._checkEL(cm.ctree, greens_tree[1].currents['L'][1])
        with pytest.raises(AssertionError):
            self._checkEL(cm.ctree, self.tree[1].currents['L'][1])
        with pytest.raises(AssertionError):
            self._checkPasCondProps(ctree_leak, ctree_all)

    def testRecalcImpedanceMatrix(self, g_inp=np.linspace(0., 0.01, 20)):
        self.loadBall()
        fit_locs = [(1, .5)]
        cm = CompartmentFitter(self.tree)
        cm.setCTree(fit_locs)

        # test only leak
        # compute impedances explicitly
        greens_tree = cm.createTreeGF(channel_names=[])
        greens_tree.setEEq(-75.)
        greens_tree.setImpedancesInTree()
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs,
                                                explicit_method=False)[0]
        z_test = z_mat[:, :,
                       None] / (1. + z_mat[:, :, None] * g_inp[None, None, :])
        # compute impedances with compartmentfitter function
        z_calc = np.array([ \
                           cm.recalcImpedanceMatrix('fit locs', [g_i], \
                               channel_names=[]
                           ) \
                           for g_i in g_inp \
                          ])
        z_calc = np.swapaxes(z_calc, 0, 2)
        assert np.allclose(z_calc, z_test)

        # test with z based on all channels (passive)
        # compute impedances explicitly
        greens_tree = cm.createTreeGF(
            channel_names=list(cm.tree.channel_storage.keys()))
        greens_tree.setEEq(-75.)
        greens_tree.setImpedancesInTree()
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs,
                                                explicit_method=False)[0]
        z_test = z_mat[:, :,
                       None] / (1. + z_mat[:, :, None] * g_inp[None, None, :])
        # compute impedances with compartmentfitter function
        z_calc = np.array([ \
                           cm.recalcImpedanceMatrix('fit locs', [g_i], \
                               channel_names=list(cm.tree.channel_storage.keys())) \
                           for g_i in g_inp \
                          ])
        z_calc = np.swapaxes(z_calc, 0, 2)
        assert np.allclose(z_calc, z_test)

    def testSynRescale(self, g_inp=np.linspace(0., 0.01, 20)):
        e_rev, v_eq = 0., -75.
        self.loadBallAndStick()
        fit_locs = [(4, .7)]
        syn_locs = [(4, 1.)]
        cm = CompartmentFitter(self.tree)
        cm.setCTree(fit_locs)
        # compute impedance matrix
        greens_tree = cm.createTreeGF(channel_names=[])
        greens_tree.setEEq(-75.)
        greens_tree.setImpedancesInTree()
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs + syn_locs)[0]
        # analytical synapse scale factors
        beta_calc = 1. / (1. + (z_mat[1, 1] - z_mat[0, 0]) * g_inp)
        beta_full = z_mat[0,1] / z_mat[0,0] * (e_rev - v_eq) / \
                    ((1. + (z_mat[1,1] - z_mat[0,0]) * g_inp ) * (e_rev - v_eq))
        # synapse scale factors from compartment fitter
        beta_cm = np.array([cm.fitSynRescale(fit_locs, syn_locs, [0], [g_i], e_revs=[0.])[0] \
                            for g_i in g_inp])
        assert np.allclose(beta_calc, beta_cm, atol=.020)
        assert np.allclose(beta_full, beta_cm, atol=.015)

    def fitBall(self):
        self.loadBall()
        freqs = np.array([0.])
        locs = [(1, 0.5)]
        e_eqs = [-75., -55., -35., -15.]
        # create compartment tree
        ctree = self.tree.createCompartmentTree(locs)
        ctree.addCurrent(channelcollection.Na_Ta(), 50.)
        ctree.addCurrent(channelcollection.Kv3_1(), -85.)

        # create tree with only leak
        greens_tree_pas = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_pas[1].currents = {'L': greens_tree_pas[1].currents['L']}
        greens_tree_pas.setCompTree()
        greens_tree_pas.setImpedance(freqs)
        # compute the passive impedance matrix
        z_mat_pas = greens_tree_pas.calcImpedanceMatrix(locs)[0]

        # create tree with only potassium
        greens_tree_k = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_k[1].currents = {key: val for key, val in greens_tree_k[1].currents.items() \
                                               if key != 'Na_Ta'}
        # compute potassium impedance matrices
        z_mats_k = []
        for e_eq in e_eqs:
            greens_tree_k.setEEq(e_eq)
            greens_tree_k.setCompTree()
            greens_tree_k.setImpedance(freqs)
            z_mats_k.append(greens_tree_k.calcImpedanceMatrix(locs))

        # create tree with only sodium
        greens_tree_na = self.tree.__copy__(new_tree=GreensTree())
        greens_tree_na[1].currents = {key: val for key, val in greens_tree_na[1].currents.items() \
                                               if key != 'Kv3_1'}
        # create state variable expansion points
        svs = []
        e_eqs_ = []
        na_chan = greens_tree_na.channel_storage['Na_Ta']
        for e_eq1 in e_eqs:
            sv1 = na_chan.computeVarinf(e_eq1)
            for e_eq2 in e_eqs:
                e_eqs_.append(e_eq2)
                sv2 = na_chan.computeVarinf(e_eq2)
                svs.append({'m': sv2['m'], 'h': sv1['h']})

        # compute sodium impedance matrices
        z_mats_na = []
        for ii, sv in enumerate(svs):
            greens_tree_na.setEEq(e_eqs[ii % len(e_eqs)])
            greens_tree_na[1].setExpansionPoint('Na_Ta', sv)
            greens_tree_na.setCompTree()
            greens_tree_na.setImpedance(freqs)
            z_mats_na.append(greens_tree_na.calcImpedanceMatrix(locs))

        # passive fit
        ctree.computeGMC(z_mat_pas)
        # get SOV constants for capacitance fit
        sov_tree = greens_tree_pas.__copy__(new_tree=SOVTree())
        sov_tree.setCompTree()
        sov_tree.calcSOVEquations()
        alphas, phimat, importance = sov_tree.getImportantModes(
            locarg=locs,
            sort_type='importance',
            eps=1e-12,
            return_importance=True)
        # fit the capacitances from SOV time-scales
        ctree.computeC(-alphas[0:1].real * 1e3,
                       phimat[0:1, :].real,
                       weights=importance[0:1])

        # potassium channel fit
        for z_mat_k, e_eq in zip(z_mats_k, e_eqs):
            ctree.computeGSingleChanFromImpedance('Kv3_1',
                                                  z_mat_k,
                                                  e_eq,
                                                  freqs,
                                                  other_channel_names=['L'])
        ctree.runFit()
        # sodium channel fit
        for z_mat_na, e_eq, sv in zip(z_mats_na, e_eqs_, svs):
            ctree.computeGSingleChanFromImpedance('Na_Ta',
                                                  z_mat_na,
                                                  e_eq,
                                                  freqs,
                                                  sv=sv,
                                                  other_channel_names=['L'])
        ctree.runFit()

        ctree.setEEq(-75.)
        ctree.removeExpansionPoints()
        ctree.fitEL()

        self.ctree = ctree

    def testFitModel(self):
        self.loadTTree()
        fit_locs = [(1, .5), (4, 1.), (5, .5), (8, .5)]

        # fit a tree directly from CompartmentTree
        greens_tree = self.tree.__copy__(new_tree=GreensTree())
        greens_tree.setCompTree()
        freqs = np.array([0.])
        greens_tree.setImpedance(freqs)
        z_mat = greens_tree.calcImpedanceMatrix(fit_locs)[0]
        ctree = greens_tree.createCompartmentTree(fit_locs)
        ctree.computeGMC(z_mat)
        sov_tree = self.tree.__copy__(new_tree=SOVTree())
        sov_tree.calcSOVEquations()
        alphas, phimat = sov_tree.getImportantModes(locarg=fit_locs)
        ctree.computeC(-alphas[0:1].real * 1e3, phimat[0:1, :].real)

        # fit a tree with compartmentfitter
        cm = CompartmentFitter(self.tree)
        ctree_cm = cm.fitModel(fit_locs)

        # compare the two trees
        self._checkPasCondProps(ctree_cm, ctree)
        self._checkPasCaProps(ctree_cm, ctree)
        self._checkEL(ctree_cm, -75.)

        # check active channel
        self.fitBall()
        locs = [(1, 0.5)]
        cm = CompartmentFitter(self.tree)
        ctree_cm_1 = cm.fitModel(locs,
                                 parallel=False,
                                 use_all_channels_for_passive=False)
        ctree_cm_2 = cm.fitModel(locs,
                                 parallel=False,
                                 use_all_channels_for_passive=True)

        self._checkAllCurrProps(self.ctree, ctree_cm_1)
        self._checkAllCurrProps(self.ctree, ctree_cm_2)

    def testPickling(self):
        self.loadBall()

        # of PhysTree
        ss = pickle.dumps(self.tree)
        pt_ = pickle.loads(ss)
        self._checkPhysTrees(self.tree, pt_)

        # of GreensTree
        greens_tree = self.tree.__copy__(new_tree=GreensTree())
        greens_tree.setCompTree()
        freqs = np.array([0.])
        greens_tree.setImpedance(freqs)

        ss = dill.dumps(greens_tree)
        gt_ = dill.loads(ss)
        self._checkPhysTrees(greens_tree, gt_)

        # of SOVTree
        sov_tree = self.tree.__copy__(new_tree=SOVTree())
        sov_tree.calcSOVEquations()

        # fails with pickle (lambda functions)
        with pytest.raises(AttributeError):
            ss = pickle.dumps(sov_tree)

        # works with dill
        ss = dill.dumps(sov_tree)
        st_ = dill.loads(ss)
        self._checkPhysTrees(sov_tree, st_)

    def testParallel(self, w_benchmark=False):
        self.loadTSegmentTree()
        locs = [(nn.index, 0.5) for nn in self.tree.nodes[:30]]
        cm = CompartmentFitter(self.tree)

        ctree_cm = cm.fitModel(locs,
                               parallel=False,
                               use_all_channels_for_passive=True)

        if w_benchmark:
            from timeit import default_timer as timer
            t0 = timer()
            cm.fitChannels(recompute=False, pprint=False, parallel=False)
            t1 = timer()
            print('Not parallel: %.8f s' % (t1 - t0))
            t0 = timer()
            cm.fitChannels(recompute=False, pprint=False, parallel=True)
            t1 = timer()
            print('Parallel: %.8f s' % (t1 - t0))
コード例 #14
0
ファイル: L23_pyramid.py プロジェクト: unibe-cns/NEAT
def getL23Pyramid():
    """
    Return a model of the L2/3 pyramidal cell with somatic and basal Na-, K- and
    Ca-channels
    """
    cm = 1.  # uF / cm^2
    rm = 10000.  # Ohm * cm^2
    ri = 150.  # Ohm * cm
    el = -75.  # mV
    tadj = 3.21

    phys_tree = PhysTree('models/morphologies/L23PyrBranco.swc')

    phys_tree.setPhysiology(
        cm,  # Cm [uF/cm^2]
        ri / 1e6,  # Ra[MOhm*cm]
    )

    # channels
    Na = channels_branco.Na()
    K_v = channels_branco.K_v()
    K_m = channels_branco.K_m()
    K_ca = channels_branco.K_ca()
    Ca_H = channels_branco.Ca_H()
    Ca_T = channels_branco.Ca_T()

    # somatic channels
    phys_tree.addCurrent(Na, tadj * 1500. * 1e2, 60.,
                         node_arg=[phys_tree[1]])  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_v, tadj * 200. * 1e2, -90.,
                         node_arg=[phys_tree[1]])  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_m, tadj * 2.2 * 1e2, -90.,
                         node_arg=[phys_tree[1]])  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_ca, tadj * 2.5 * 1e2, -90.,
                         node_arg=[phys_tree[1]])  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(Ca_H, tadj * 0.5 * 1e2, 140.,
                         node_arg=[phys_tree[1]])  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(Ca_T, 0.0003 * 1e6, 60.,
                         node_arg=[phys_tree[1]])  # mho/cm^2 -> uS/cm^2
    phys_tree.setLeakCurrent(1. / rm * 1e6, el, node_arg=[phys_tree[1]])

    # basal channels
    phys_tree.addCurrent(Na, tadj * 40. * 1e2, 60.,
                         node_arg='basal')  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_v, tadj * 30. * 1e2, -90.,
                         node_arg='basal')  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_m, tadj * 0.05 * 1e2, -90.,
                         node_arg='basal')  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(K_ca, tadj * 2.5 * 1e2, -90.,
                         node_arg='basal')  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(Ca_H, tadj * 0.5 * 1e2, 140.,
                         node_arg='basal')  # pS/um^2 -> uS/cm^2
    phys_tree.addCurrent(Ca_T, 0.0006 * 1e6, 60.,
                         node_arg='basal')  # mho/cm^2 -> mho/cm^2
    phys_tree.setLeakCurrent(1. / rm * 1e6, el, node_arg='basal')

    # passive apical dendrite
    phys_tree.setLeakCurrent(1. / rm * 1e6, el, node_arg='apical')

    return phys_tree
コード例 #15
0
ファイル: L5_pyramid.py プロジェクト: unibe-cns/NEAT
def getL5Pyramid():
    """
    Return a minimal model of the L5 pyramid for BAC-firing
    """
    # load the morphology
    phys_tree = PhysTree('models/morphologies/cell1_simplified.swc')

    # set specific membrane capacitance and axial resistance
    phys_tree.setPhysiology(1., # Cm [uF/cm^2]
                            100./1e6, # Ra[MOhm*cm]
                            node_arg=[phys_tree[1]])
    # set specific membrane capacitance and axial resistance
    phys_tree.setPhysiology(2., # Cm [uF/cm^2]
                            100./1e6, # Ra[MOhm*cm]
                            node_arg=[n for n in phys_tree if not phys_tree.isRoot(n)])

    # channels present in tree
    Kv3_1  = channels_hay.Kv3_1()
    Na_Ta  = channels_hay.Na_Ta()
    Ca_LVA = channels_hay.Ca_LVA()
    Ca_HVA = channels_hay.Ca_HVA()
    h_HAY  = channels_hay.h_HAY()

    # soma ion channels [uS/cm^2]
    phys_tree.addCurrent(Kv3_1,  0.766    *1e6, -85., node_arg=[phys_tree[1]])
    phys_tree.addCurrent(Na_Ta,  1.71     *1e6,  50., node_arg=[phys_tree[1]])
    phys_tree.addCurrent(Ca_LVA, 0.00432  *1e6,  50., node_arg=[phys_tree[1]])
    phys_tree.addCurrent(Ca_HVA, 0.000567 *1e6,  50., node_arg=[phys_tree[1]])
    phys_tree.addCurrent(h_HAY,  0.0002   *1e6, -45., node_arg=[phys_tree[1]])
    phys_tree.setLeakCurrent(0.0000344 *1e6, -90., node_arg=[phys_tree[1]])

    # basal ion channels [uS/cm^2]
    phys_tree.addCurrent(h_HAY, 0.0002 *1e6, -45., node_arg='basal')
    phys_tree.setLeakCurrent(0.0000535 *1e6, -90., node_arg='basal')

    # apical ion channels [uS/cm^2]
    phys_tree.addCurrent(Kv3_1, 0.000298 *1e6, -85., node_arg='apical')
    phys_tree.addCurrent(Na_Ta, 0.0211   *1e6,  50., node_arg='apical')
    phys_tree.addCurrent(Ca_LVA, lambda x: 0.0198*1e6   if (x>685. and x<885.) else 0.0198*1e-2*1e6,   50., node_arg='apical')
    phys_tree.addCurrent(Ca_HVA, lambda x: 0.000437*1e6 if (x>685. and x<885.) else 0.000437*1e-1*1e6, 50., node_arg='apical')
    phys_tree.addCurrent(h_HAY,  lambda x: 0.0002*1e6 * (-0.8696 + 2.0870 * np.exp(x/323.)),          -45., node_arg='apical')
    phys_tree.setLeakCurrent(0.0000447*1e6, -90., node_arg='apical')

    return phys_tree