DeepPurpose: An AI Toolkit for Accelerating COVID‑19 Drug Discovery

56 Cutting‑Edge Models, Full Features

DeepPurpose consists of two encoders that generate embeddings for drug molecules and proteins, which are then concatenated and fed into a decoder to predict binding affinity between a drug‑target pair.

The input is a drug‑target pair and the output is a score indicating their binding activity.

Both drug and protein encoders come in multiple types: eight encoders for molecules and seven for proteins, yielding 7 × 8 = 56 possible model combinations, many of which are state‑of‑the‑art.

Get Started in Under 10 Steps

The entire workflow can be completed in fewer than ten steps, each typically requiring only one line of code:

1. Data loading 2. Specify encoder 3. Split and encode dataset 4. Generate model configuration 5. Initialize model 6. Train model 7. Old‑drug repurposing / virtual screening 8. Save / load model

After training, DeepPurpose can automatically generate affinity scores for drug‑target pairs, rank them, and support both drug repurposing and virtual screening tasks.

The toolkit also includes the MIT‑collected open COVID‑19 dataset, with ready‑to‑use functions for loading and processing the data.

Target Protein: Drug’s Action Object

Drug discovery fundamentally relies on assessing the affinity between a drug molecule and its target protein; many diseases are linked to over‑expressed or malfunctioning proteins, making them ideal therapeutic targets.

AI Boosts New‑Drug R&D

Traditional drug development can take around 15 years, with the research‑development phase alone consuming 2–10 years due to extensive experimental screening.

Applying AI to predict drug‑target interactions can dramatically reduce this timeline by automating the screening process and focusing experimental effort on the most promising candidates.

Author Introduction

The first author, Huang Kexin, holds dual bachelor's degrees in mathematics and computer science from NYU and is pursuing a master's at Harvard focusing on medical big data. His research centers on graph neural networks (GNN) for drug discovery and medical text mining.

Co‑authors Tianfan Fu, Lucas Glass, Marinka Zitnik, Cao Xiao, and Jimeng Sun also contributed to the study.

For further reading, scan the QR code below to receive a free Python course and additional learning resources.