intro.py

# %% [markdown]
## chModeler

# A new approach to the old Hodgkin-Huxley model for: Making all ion channel models comparable.
#
# This package builds models for ion channels from patch clamp data using Binomial Distribution.
#
# The default dataset contains ~200 voltage-gated ion channel data from different species with similar experiment conditions where data have been gathered from literature and digitized for modeling and comparision studies.
#
# This dataset is temporarily available at: http://chopen.herokuapp.com 


## Ion Channel Modeling Using chModeler and Binomial Distribution
#* **Problems with HH-style models:**
#  * Independent m and h when having both activation and inactivation which is in contrast with dependent nature activation and inactivation gating of ion channels  
#  * No unified equation expressions
#    * Different alpha-beta expressions in different models
#    * Different inactivation expressions
#  * Too many parameters for many models  
#  * Parameters are varied from one model to the other
#  * Many HH models can't describe kinetics very well at different voltages (esp. low voltages)
#    * Different powers at different voltages!!
#* **Fitting problems:**
#  * Finding gmax is so tedious when we have both activation and inactivation
#  * Parameter estimation for tau, alpha, and beta is usually not unique when having both activation and inactivation
#  * No accurate way to decide how much of a raising current is due to activation and how much for inactivation
#  * It's usually hard to find h0 of inactivation  
#* **Markov models problems:**
#  * Complexity
#  * Number and types of states
#  * Number of parameters
#  * And the same problems mentioned above for fitting processes
#* **Binomial Distribution And ion Channel Modeling**
#  * "In probability theory and statistics, the [Binomial Distribution](https://en.wikipedia.org/wiki/Binomial_distribution) with parameters *n* and *p* is the discrete probability distribution of the number of successes in a sequence of *n* independent experiments, each asking a yes–no question, and each with its own boolean-valued outcome: success/yes/true/one (with probability *p*) or failure/no/false/zero (with probability *q = 1 − p*)."
#  
#    $$f(k,n,p) = Pr(k;n,p) = Pr(X=k) = \binom{n}{k}p^k(1-p)^{n-k}, \binom{n}{k}=\frac{n!}{k!(n-k)!}$$
#              
#  * At microscopic level, each ion channel could be only in one of two positions: either open or close. So, at macroscopic level, we can assume the number of gates in ion channels being in the activated state as a Binomial Distribution.
#  * According to the Hodgkin-Huxley equation, the current passing through all (the same) voltage-gated ion channels of a cell can be described as:
#  
#    $$I = \bar{g} * Po * (V - E_{rev})$$
#            
#  * Where, *Po* is the probability of ion channels being in open state at each time. So, we can write this probability as:
#  
#    $$Po = P_{peak} * Pr(3;4,p)$$
#    
#  * For each voltage, *P_peak* is the probability of ion channels at their relative peak conductances and *Pr* is the probability of ion channels being in activated state at each time.
#  * We can calculate the *P_peak* at each voltage using a Boltzmann function in the form of:  
#  
#    $$P_{peak} = \frac{1} { 1 + \exp(\frac{-(v - v_{half})} {slope})}$$
#    
#  * Experimentally, it has been shown that for many voltage-gated ion channels the rate of activation to inactivation is 3:1. So, we can calculate our Binomial Distribution as: 
#  
#    $$Pr = \frac{4!}{3!*1!} * p^3 * q^1$$
#    
#  * However, in order to consider both types of inactivation (fast and slow), the probability of inactivation is the product of the probability of these two types:
#  
#    $$p = 1-q, q = q1*q2$$
#    
#  * And each inactivation probability can be described as the following equations which are also being used in current ion channel modeling problems. (Please note that this probability could be 1 for any inactivation type if it's not applicable) 
#  
#    $$q1 = q1_{\infty}-((q1_{\infty}-q1_{0})*\exp(-(\alpha+\beta)*t))$$
#  
#    $$q2 = q2_{\infty}-((q2_{\infty}-q2_{0})*\exp(-(\alpha+c)*t))$$
#    
#  * *alpha* and *beta* rates can be calculated as below (and *c* which is a rate constant that also can be 0) 
#
#    $$\alpha = r_{\alpha} * \exp(-s_{\alpha}*v)$$
#
#    $$\beta = \frac{r_{\beta}} {1 + \exp(\frac{-(v - vh_{\beta})} {s_{\beta}})}$$
#
#    $$c = r_{c}$$
#  
#* **Fitting Data Using chModeler**
#  * chModeler solves Independent m and h problem (we have p and 1-p instead)
#  * We have only one equation and 10 parameters for all VG ion channel models
#  * The new algorithm solves the fitting problems
#  * Initial guesses could be both blind and from previous models in DataSet
#  * In blind guess method it takes a sec for fitting a model
#  * Using previous models a more accurate model could be built in a minute

# %%
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
image1 = mpimg.imread("doc/bd.png")
plt.imshow(image1)
plt.title("Binomial Distribution")
plt.show()
image2 = mpimg.imread("doc/ic5.png")
plt.imshow(image2)
plt.show()

# %% [markdown]
# ## Installation

#In order to install and run this project within current notebook run the following cell (make sure you are using python >= 3.4):
# %%
import sys;
!{sys.executable} -m pip install .;

# %% [markdown]
# ## Usage
#You can run `python3 chModeler.py -h` for command-line usage help.
# %%
import sys
!{sys.executable} chModeler.py -h

# %% [markdown]
# **There are 3 ways to initiate running chModeler:**
#
#* **Sending command arguments to chModeler directly:** 
#
# Build model for ion channel No.5 in dataset and save the model in /data directory using default options.
#
# `python3 chModeler.py -i 5`
# %%
%%time
!{sys.executable} chModeler.py -i 5

# %% [markdown]
# Do not save results and show plots for each fitting steps (True/1 and False/0 can be used alternatively) 
# %%
%%time
!{sys.executable} chModeler.py -i 5 -s False -p True

# %% [markdown]
# Use previous models to fit the model with r2 score threshold of 0.98 (increase r2 (e.g. 0.999) for a smaller but faster initial state.) 
#
# `python3 chModeler.py -i 5 -ft 2 -r2 0.98`
# %%
%%time
!{sys.executable} chModeler.py -i 5 -ft 2 -r2 0.98

# %% [markdown]
# Use the model located in `data/2ndFit/18_Sh-B1_DROME_10p_2.json` to fit and plot only final results 
#
# `python3 chModeler.py -i 5 -ft 2 -s 0 -fp 1 -mf "data\2ndFit\18_Sh-B1_DROME_10p_2.json"`
# %%
%%time
!{sys.executable} chModeler.py -i 5 -ft 2 -s 0 -fp 1 -mf "data\2ndFit\18_Sh-B1_DROME_10p_2.json"

# %% [markdown]   
# * **Reading command arguments from a file:**
#  
# Read the arguments from args.txt file
# %%
%%time
!{sys.executable} chModeler.py "@args.txt"

# %% [markdown]
# * **Running the wizard for answering questions and entering the options:**
#
# Run the wizard
#
# `python3 chModeler.py -w` 
# %%
# !{sys.executable} chModeler.py -w


# %% [markdown]
# ## TODOs:

# * **Model Validation**: Because these models are based on digitized data from literature, voltage traces are dependent to the related experiment. So. in order to verify each model to a wide range of voltages there is a need to validate final models against a wider range of voltages.
# * **Smarter Initial Parameter Selection**: Right now, previous models for next fittings are being selected based on R2 scores. This can be improved to select models from the same category using some classification technique.
# * **Temperature and Calcium concentration dependency**: Although the *c* constant can cover the calcium concentration dependency, it's needed to be validated.
# * **Standard data support**: This package needs to be updated to support data analysis from patch clamp machines.