PaddleHelix

PaddleHelix is Baidu’s open-source bio-computing platform built on PaddlePaddle, featuring large-scale representation learning and multi-task deep learning for molecules and proteins, including the HelixFold family of biomolecular structure prediction models. The project has about 1,100 stars on GitHub and is actively developed in 2026.

Overview

Drug discovery and protein science increasingly run on learned representations rather than hand-crafted features. PaddleHelix packages that capability into one platform: pre-trained models for compounds and proteins, and toolkits for the tasks researchers actually run, including molecular property prediction, drug-target interaction, drug-drug interaction, protein-protein interaction, protein folding and docking, molecule design and RNA structure prediction. Its flagship component, HelixFold, predicts biomolecular structures in the AlphaFold tradition; the newer HelixFold3 line targets complexes and improved structural quality, with versions released through 2025. The platform emphasizes self-supervised pre-training on large-scale unlabeled biomedical data, then fine-tuning for specific tasks, all within the PaddlePaddle ecosystem that is widely deployed in Chinese industry and academia.

Key Features

• Bio-computing platform for molecules, proteins and RNA
• HelixFold structure prediction, up to the HelixFold3 generation
• Pre-trained representation models for compounds and proteins
• Toolkits for DTI, DDI, PPI, docking and molecule design tasks
• Multi-task and self-supervised learning pipelines included
• Built on the PaddlePaddle deep learning framework
• Actively maintained by Baidu’s PaddlePaddle team

PaddleHelix also serves as the home of the HelixFold series’ released code and parameters, which drew wide attention in the AI-for-science community as an open counterpart to commercial structure prediction services. Its graph neural network toolkits for molecular representation learning are used independently of the folding models, for example in molecular property prediction benchmarks and screening pipelines.

💳 Pricing

The PaddleHelix code and model parameters are free and open source on GitHub for self-hosted research. Baidu also offers a paid API for convenient, high-throughput access to HelixFold3 predictions for academic and commercial users, making the platform freemium overall.

Who Is It For

PaddleHelix is for computational biologists, pharmaceutical researchers and AI-for-science teams working on drug discovery, protein structure prediction and molecular property modeling.