yenanlai / caenorhabditis-elegans Goto Github PK

This repository contains the source code used in the dissertation Behavioural Correspondence of Neuronal Dynamics in Caenorhabditis elegans.

The research dataset employed in this dessrtation, Kato2015 whole brain imaging data, was acquired from the OSF open repository. This invaluable resource, made accessible by Manuel Zimmer, can be accessed via the following URL: https://osf.io/2395t/?view_only=.

Opt for two datasets, namely, WT_NoStim.mat and WT_Stim.mat.

• WT_NoStim dataset: Signifying the wild type with the absence of sensory stimulation.

• WT_Stim dataset: Denoting the wild type wherein oxygen chemosensory neurons are activated via consecutive oxygen upshifts and downshifts (21% as opposed to 4%).

In this dissertation, two machine learning models, namely the Logistic Regression Classifier and the Random Forest Classifier, were chosen for analysis. These models were systematically trained and evaluated using two distinct datasets: WT_NoStim dataset and WT_Stim dataset.

• Train the Logistic Regression Classifier on the WT_NoStim dataset.
• Train the Random Forest Classifier on the WT_NoStim dataset.
• Train the Logistic Regression Classifier on the WT_Stim dataset.
• Train the Random Forest Classifier on the WT_Stim dataset.

Jupyter Notebook used in this dissertation is stored in the directory notebook.

The notebook delineates the comprehensive workflow, encompassing stages of data preprocessing, model formulation and training, model validation, cross-validation procedures, hyperparameter optimization, among other processes.

Packages in this repository are available in the directory src.

• model.py Definition of Logistic Regression Classifier and Random Forest Classifier are established.
• wt_nostim_data_preprocessing.py The WT_NoStim dataset undergoes preprocessing and is subsequently partitioned into training and test sets, adhering to a 7:3 ratio.
• wt_nostim_train_validation.py Logistic Regression Classifier and Random Forest Classifier are trained and evaluated utilizing the WT_NoStim dataset.
• wt_stim_data_preprocessing.py The WT_Stim dataset undergoes preprocessing and is subsequently partitioned into training and test sets, adhering to a 7:3 ratio.
• wt_stim_train_validation.py Logistic Regression Classifier and Random Forest Classifier are trained and evaluated utilizing the WT_Stim dataset.
• cross_validation.py Average cross-validation scores for both the Logistic Regression Classifier and Random Forest Classifier are computed, considering both the entire WT_NoStim dataset and the entire WT_Stim dataset.
• feature_importance.py Some functions are established for quantifying the significance of neuronal features.