# Drug Design Challenge — Data

来源：https://www.kaggle.com/competitions/drug-design-challenge/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

## Dataset 
To achieve the objective explained in the previous section, a machine learning model should be trained to be used for virtual screening of chemical libraries such as ZINC, CHEMBLE. 
The present dataset includes the human ErbB1 target and an IC50 activity (enzyme inhibition parameter). The dataset contains 6953 molecules divided into 5215 molecules for training and 1738 for testing. Each molecule is represented using 1416 molecular descriptors. 
In this challenge, the following column is targeted:

1.	Regression target (pIC50_nM): continuous target which should be predicted using regression modeling. Here, the R^2 should be used as a performance measure. 

It is preferable that the work be done in an ipython notebook (Google collab could be considered as runtime). Moreover, all data science phases should be considered (Exploratory analysis, preprocessing, visualizations, modeling, model evaluation).

## Useful links
●	[The half maximal inhibitory concentration (IC50)](https://en.wikipedia.org/wiki/IC50#:~:text=IC50%20is%20a%20quantitative,cell%2C%20cell%20receptor%20or%20microorganism.)
●	[Quantitative structure–activity relationship models (QSAR)](https://en.wikipedia.org/wiki/Quantitative_structure%E2%80%93activity_relationship)
●	[ErbB1](https://www.cancer.gov/publications/dictionaries/cancer-terms/def/erbb1)