BGA-Classification

Overview

This repository contains code to classify individuals based on an AIM set into populations. The code was built with TabPFN, which has been published in Nature "[Accurate predictions on small data with a tabular foundation model] by Hollmann et al.

This repsository contains:

• Application_TabPFN.py: to use TabPFN for classification of individuals into populations that can be downloaded. You just have to change the input paths to your test and training data (Application_TabPFN.py).
• CrossValidation_Code/run_experiments.py: a python script that was used to run our experiments.
• Streamlit_TabPFN.py: a Grafical User Interface to use TabPFN for classification of individuals into populations, which runs in the Browser.
• data_test.csv, data_train.csv: Example test and training data. This is a subset of the data published as a Supplemental (1-s2.0-S1872497323000285-mmc5.xlsx) of Ruiz-Ramirez et al.

• prediction_prob.csv: Example output prediction probabilites.
  

• Extract_EUROPEAN.py, Extract_continents.py: the code that was used to prepare the data of our paper Enhancing BGA Prediction with machine learning.
• Cross_Validation_Code/Confusion_matrix.py: the code to plot the confusion matrices.
• Cross_Validation_Code/run_plotting.py: python code to plot the mean of ROC AUC, accuracy and logloss (or any other metric) and its confidence interval.
• Cross_Validation_Code/SNIPPER_CrossValidation.R: Code for the Cross Validation with SNIPPER
• Cross_Validation_Code/PLS_DA_CrossValidation.py: Code for the Cross Validation with PLS-DA
• Cross_Validation_Code/Admixture_Model_CrossValidation.py: Code for the Cross Validation with the Admixture Model

Usage: Web Tool

See here for an online version of the Grapical User Interface. Note that there are no data protection rules on the server where this interface runs.

Usage: Local Graphical User Interface

To use the graphical user interface, download the repository and install the requirements using

pip install -r requirements.txt

Afterwards, type

streamlit run Streamlit_TabPFN.py

Then, your browser should open and you can use the example test and trainings data or you can upload your own data.

Usage: Command Line

With

python Application_TabPFN.py

a TabPFN run is performed on train_data_eur.csv and test_data_eur.csv. Change the file according to your needs if it should run on a different dataset.