PineBioML 1.2.0

PineBioML is a easy use ML toolkit.

Overview

This package aims to help analysising biomedical data using ML method in python.

System requirements

1. Python 3.10+
2. The following python module dependencies are required:

pandas openpyxl xlrd tqdm seaborn gprofiler-official jupyter jupyterlab optuna scikit-learn umap-learn pacmap statsmodels mljar-supervised joblib

Installation

1. Install Python

Please follow the tutorial to install python (the sections "Visual Studio Code" and "Git" are optional):

https://learn.microsoft.com/en-us/windows/python/beginners 

Please skip this step if you already have python 3.10+ installed in your PC.

2. Install dependencies and execute the scripts

Step 1. Download our scripts from Release and unzip it.

https://github.com/ICMOL/undetermined/releases

Step 2. Install dependencies: Please open Windows PowerShell, move to the directory of our scripts, and execute the following command.

pip install -r ./requirements.txt          

Step 3. Open the jupyter interface

Please execute the following command to open jupyter. You will see the figure, if the scripts execute correctly.

> jupyter lab    

or

> python -m notebook

Input Table Format

The input data should be tabular and placed in the ./input folder. We accept .csv, .tsv, .xlsx and R-table in .txt formats.

Process

0. Document

API

1. Missing value preprocess

ID	Option	Definition
1	Deletion	Remove the features that are too empty.
2	Imputation with a constant value	Impute missing values with a constant value, such as 0 or the feature mean.
3	Imputation using K-NN algorithm	Impute missing values with the mean or median of the k nearest samples.

2. Data transformation

ID	Option	Definition
1	PCA	Principal component transform.
2	Power transform	To make data more Gaussian-like, you can use either Box-Cox transform or Yeo-Johnson transform.
3	Feature clustering	Group similar features into a cluster.
4	Feature expansion	Generating new features by add/product/ratio in random pair of existing features.

3. Feature selection

ID	Option	Definition
1	Volcano plot	Selecting by group p-value and fold change
2	Lasso regression	Selecting by Linear models with L1 penalty
3	Decision stump	Selecting by 1-layer decision tree
4	Random Forest	Selecting by Gini impurity or permutation importance over a Random Forest
5	AdaBoost	Selecting by Gini impurity over a AdaBoost model
6	Gradient boosting	Selecting by Gini impurity over a gradient boosting, such as XGboost or LightGBM
7	Linear SVM	Selecting by support vector from support vector machine

4. Model building

ID	Option	Definition
1	ElasticNet	Using Optuna to find a not-bad hyper parameters on given dataset.
2	SVM	Using Optuna to find a not-bad hyper parameters on given dataset.
3	Decision Tree	Using Optuna to find a not-bad hyper parameters on given dataset.
4	Random Forest	Using Optuna to find a not-bad hyper parameters on given dataset.
5	AdaBoost	Using Optuna to find a not-bad hyper parameters on given dataset.
6	XGBoost	Using Optuna to find a not-bad hyper parameters on given dataset.
7	LightGBM	Using Optuna to find a not-bad hyper parameters on given dataset.
8	CatBoost	Using Optuna to find a not-bad hyper parameters on given dataset.

5. Report and visualization

ID	Option	Definition
1	data_overview	Giving a glance to input data.
2	classification_summary	Summarizing a classification task

Examples for Program Demonstration

Chosse one of the following examples, double click it in jupyter interface:

ID	Name	Description
1	example_BasicUsage.ipynb	Demonstrate the basic features of PineBioML
2	example_Proteomics.ipynb	An example on proteomics data analysis
3	example_PipeLine.ipynb	Demonstrate how to use the pipeline to store the whole data processing flow
4	example_Pine.ipynb	Demonstrate how to use Pine ml to finding the best data processing flow in an efficient way
5	example_UsingExistingModel.ipynb	An example of unsing existing models/pipeline gained from 3. , 4. or 5.

Click the buttom and the script should start.

Cites

The example data is from LinkedOmicsKB

A proteogenomics data-driven knowledge base of human cancer, Yuxing Liao, Sara R. Savage, Yongchao Dou, Zhiao Shi, Xinpei Yi, Wen Jiang, Jonathan T. Lei, Bing Zhang, Cell Systems, 2023.