---
title: DeepPurpose
description: Train and run DeepPurpose models for drug-target prediction on the platform.
tags:
  - deeppurpose
---

DeepPurpose predicts drug–target interactions, adding a kinome-wide off-target read that a general ADMET panel does not cover. Reach for it when selectivity is the question — checking a compound against a panel of kinase targets during lead optimization, or profiling off-target liabilities before committing to a series. It is narrower than a broad ADMET screen: pair it with [ADMET-AI](/docs/admet-ai) for the fast safety panel. See [what you can achieve](/docs/capabilities#admet-and-safety) for the full set.

DeepPurpose provides ready-to-use encoders for SMILES and protein sequences plus
tools for drug–target, drug–drug, and side-effect predictions. Use these guides
to stand up repeatable jobs for both research and production scoring.

## Citations

> Huang K, Fu T, Glass LM, Zitnik M, Xiao C, Sun J. DeepPurpose: a deep learning library for drug–target interaction prediction. *Bioinformatics*. 2021;36(22-23):5545-5547. doi: [10.1093/bioinformatics/btaa1005](https://doi.org/10.1093/bioinformatics/btaa1005)

## Chat Usage

### Example Prompts

- "Predict kinome-wide off-targets for sorafenib using DeepPurpose"
- "Run DeepPurpose DTI prediction for gefitinib against EGFR"
- "Screen my compound against a panel of kinase targets"

### Workflow Chaining

- **Feeds into**: ADMET-AI (safety for predicted binders), Target Passport (off-target analysis)
- **Accepts from**: PubChem (compound SMILES), UniProt (target sequences), REINVENT (generated molecules)

## FAQ

### How do I install DeepPurpose?

  Use pip with GPU-enabled dependencies and pin CUDA-compatible torch; see the quickstart.

### Can I use the model zoo?

  Yes. Load pretrained encoders/decoders from the model zoo and evaluate before fine-tuning.

### How do I export predictions?

  Save CSV outputs with molecule/target IDs, scores, and metadata; optionally export ONNX for serving.

## Read next