20k+ Proteins

2M+ Publications

2k+ Druggable Proteins

600+ Investigational Proteins

Select Method:

Latent Features ESM Embeddings

A total of 183 features have been used for predicting the druggability index, in doing so, some features have been shown to be of more importance in terms of their contribution. Here, we show the contribution of each of the protein-features.

We have used Random Forest and XGBoost models to accurate predict the druggability potential of a human protein. Users are free to download the results mapping the DI values obtained for each protein.

FAQ's

What does DrugProtAI do?

DrugProtAI is a comprehensive web-based tool designed for evaluating the druggability of proteins. It allows users to search for a protein using its UniProt ID and provides detailed information, including feature contributions from machine learning models like XGBoost (XGB) and Random Forest (RF), druggability predictions, and an interactive 3D protein structure viewer. The platform integrates data from several knowledgebases such as UniProt, DrugBank, PubMed, and Alphafold, making it a powerful resource for researchers aiming to assess the potential of proteins as drug targets.

The Druggability Index (DI) of a protein is then defined as the probability of the given protein being approved-druggable

What is the meaning of druggability index?

What are the features on which the tool has been developed?

DrugProtAI is built using 183 features from UniProt, including protein domains, PPI properties, PTM counts, glycosylation, subcellular locations, flexibility sequence properties, and physicochemical properties. Additional features include latent values from an autoencoder and GDPC encodings. These features capture diverse biological and physicochemical aspects critical for predicting protein druggability.

DrugProtAI uses two machine learning models: XGBoost (XGB) and Random Forest (RF). These models are trained on 183 protein features, including biological, physicochemical, and structural properties. The reliability of these models is ensured through rigorous cross-validation and testing, with feature contributions visualized via pie charts. These models are widely used in bioinformatics for their accuracy and robustness in classification tasks like druggability prediction.

What model is employed to get the result? How reliable is the model?

Does it work on any protein or specific to a species?

DrugProtAI is primarily designed to work with human proteins, as its dataset consists of 20,273 human proteins verified by SwissProt from UniProtKB. While the tool's models and features are optimized for human proteins, extending it to other species would require additional training and dataset adjustments.