Leash-BELKA

Models to participate the Leash-BELKA competion (Ranked 122nd out of 1950 teams with Bronze medal)

Competion objectives:

Predict small molecule-protein interactions using the Big Encoded Library for Chemical Assessment

Problem definition and exploratory data analysis

Installation

General

conda create --name BELKA python=3.10
conda activate BELKA
bash install.sh

Data

• Main data: kaggle competitions download -c leash-BELKA -p

Model

Combining XGBoost model and 1D CNN model

XGBoost model:

• XGboost/scr/train_XGboost_sEH.py
• XGboost/scr/train_XGboost_BRD4.py
• XGboost/scr/train_XGboost_HSA.py

1D CNN

• 1D_CNN/scr/1D_CNN_train.py

Result

1D CNN

Submitted results

Public Leaderboard: 0.45582 (Rank 147) Private Leaderboard: 0.25429 (Rank 122)

Acknowledgement

Some code here are adapted from other solutions: (https://www.kaggle.com/code/ahmedelfazouan/belka-1dcnn-starter-with-all-data), (https://www.kaggle.com/code/gallo33henrique/model-ml-xgboost-90-accuracy-new-medicines-belka)