A deep learning library for drug-target interaction, drug property, protein-protein interaction, drug-drug interaction, and protein function prediction in bioinformatics.
DeepPurpose is a deep learning library for predicting interactions and properties of drugs and proteins. It solves key bioinformatics problems like drug-target interaction prediction, virtual screening, and drug repurposing by providing a unified toolkit with multiple neural network architectures and encodings.
Bioinformatics researchers, computational chemists, and drug discovery scientists who need to apply deep learning to molecular data without extensive coding.
Developers choose DeepPurpose for its simplicity—complex workflows require under 10 lines of code—and its comprehensive support for multiple tasks, encodings, and datasets, all built on a flexible PyTorch foundation.
A Deep Learning Toolkit for DTI, Drug Property, PPI, DDI, Protein Function Prediction (Bioinformatics)
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Supports DTI, DDI, PPI, compound property, and protein function predictions in one toolkit, eliminating the need to switch between specialized tools.
Offers 15+ drug and protein encodings including GNNs, transformers, and classic fingerprints, providing flexibility for model experimentation without custom implementation.
Enables full workflows like training and virtual screening in under 10 lines of code, as demonstrated in the one-liner repurposing examples.
Includes pre-processed data for BindingDB, DAVIS, KIBA, and COVID-19 targets, reducing time spent on tedious data preparation.
Provides over 10 checkpoints for quick inference and transfer learning, allowing users to skip initial training phases for common tasks.
The README explicitly cautions that pretrained models cover small datasets and may not generalize well to unseen proteins, necessitating custom training for real-world applications.
Installation requires a specific conda environment with multiple steps, including handling dependencies like descriptastorus, which can be error-prone compared to simpler pip installs.
Official documentation is noted as under active development, forcing users to rely on demos and tutorials that may not cover advanced or edge-case scenarios.
Encodings are limited to 1D/2D representations like SMILES and sequences, missing critical 3D geometric information needed for accurate binding predictions in some domains.