ML_project_2

Second project for the course of machine learning CS433 in collaboration with the Laboratory of Computational Systems Biotechnology (SB/SV/STI) LCSB, based on the article "Particle-based simulation reveals macromolecular crowding effects on the Michaelis-Menten mechanism", by Weilandt, Daniel R and Hatzimanikatis, Vassily.

The purpose of this project is to find a model to unravel the effects of crowding on the kinetics of biochemical reactions. The lab provided us with simulated data: the software GEneralized Elementary Kinetics (GEEK), predicts the enzymatic rate constants for varying volume exclusion conditions (i.e. the space occupied by inert molecules representative for a crowded environment), mass and mass distribution of the inert molecules, and reactive species concentrations. Starting from this simulation data, our study aims to develop a model predicting the rate constants in a black box machine learning approach.

Libraries used

We used the following libraries for this project, with Python 3.6.5

Computational:

numpy (as np)
sklearn (scikit-learn)
Statsmodel
Xgboost

Graphical:

seaborn (as sns) (version 0.9.0)
matplotlib (as plt)

Prerequisites

To install some of the libraries mentionned before , please use the following command (linux):

pip3 install xgboost
pip install -U statsmodels

The folder structure has to be the following:

.
├── Data                              # Data files, in .csv
    └── result_full_factorial_pgm.csv
├── src                               # Source files
    └── Run_general.py
├── results
    ├── plots_evolution
    └── reproduction.csv
└── README.md

Implementations details

Run.ipynb

Import functions from regressions.py, reproduction.py,figure_generation.py

Jupyter notebook in which we reproduce the different tables and figures we use in our report.

Reproduce the model described in the original article, which is a weighted linear model (see the report for more details on the weighting process).It outputs a .csv file containing the estimate of the coefficient and store it in ../results/reproduction.csv.

Plots the figure 2 of the supplementary material of the paper, which is a figure of the 4 rate constants against the 4 concentrations (plot with 16 subfigures). This figure essentially shows the influence of the input variables (concentrations) on the outcome variables (rate constants) for different volume fractions which are color-coded. The output is in ../results/Fig2_supp.png.

run_plots.py

Import function from plots.py

This script plots the prediction made by our final mdel xgboost. More precisely, having fixed all the input of our model except two concentrations, we plot the value of the kinetic constant in colour, while the axes represent the evolution of the two concentrations we chose before.

The fixed values are usually fixed as being the median of what we had in our training data (result_full_factorial_pgm.csv).

The script produces the plots for all pairs of concentrations, for each volume fraction and for each output.

It saves the results in ../results/plots_evolution, then classes the plots by the value of the volume fraction considered.

References:

Particle-based simulation reveals macromolecular crowding effects on the Michaelis-Menten mechanism
Daniel R Weilandt, Vassily Hatzimanikatis
doi: https://doi.org/10.1101/429316

Seabold, Skipper, and Josef Perktold. “Statsmodels: Econometric and statistical modeling with python.” Proceedings of the 9th Python in Science Conference. 2010.

Hastie, T., Tibshirani, R.,, Friedman, J. (2001). The Elements of Statistical Learning. New York, NY, USA: Springer New York Inc..

Authors

• William Cappelletti
• Charles Dufour
• Marie Sadler