Regression in Drug Discovery

The sample input data shown above is in CSV file format. It shows the small molecule in computer-readable format using SMILES notation. Each line includes a compound and its associated lipophilicity value.

Lipophilicity is important when evaluating whether a drug component has the potential to fail, especially if it prevents a compound from advancing through preclinical stages.

There’s a data engineering aspect involved: each compound is paired with a corresponding numeric value. The value can be positive or negative depending on the context. A negative value may indicate an undesirable compound, while a positive value may suggest one worth pursuing—or the reverse, depending on the specific goal of the model.

The main concern is creating a model that, when given unknown compounds, can accurately predict the lipophilicity.

import pandas as pd
from pathlib import Path
from lightning import pytorch as pl 
from chemprop import data, featurizers, models, nn

This code example shows the training setup in a Jupyter notebook.