🧪 Predict pIC50 Values: DrugHunter employs state-of-the-art prediction models to calculate the pIC50 values for a multitude of human protein inhibitors. This information is of paramount importance in the assessment of the potency and efficacy of potential drugs. Users may select from a range of 336 human proteins for which they wish to predict drug bioactivity. Proteins such as acetylcholinesterase and HERG can be selected according to the user's specific needs.

📙 Comprehensive Protein Database: DrugHunter includes an extensive collection of human protein structures, including known inhibitors and their associated pIC50 values. This extensive database increases the accuracy and reliability of predictions. The application provides a suite of powerful machine learning models, including Random Forest, k-Nearest Neighbours, LightGBM, and neural networks, to ensure accurate prediction of pharmacological activity.

🙋 Predictions Based on Descriptor or Fingerprint Data: Users have the flexibility to choose between molecular descriptors or Morgan fingerprints as the basis for their predictions. Both options provide reliable methods for analyzing chemical compounds. For each SMILES input provided by the user, the application generates a two-dimensional spatial structure representation of the compound. This visual aid helps users better understand the chemical structure they are working with.

📏 Identification of Similar Compounds: The application displays similar chemical compounds based on a minimum of 70% structural similarity to the compound entered by the user. This feature allows users to explore related compounds that may have similar bioactivity.

💻 User-friendly Interface: The user-friendly interface of DrugHunter, written in Streamlit, facilitates the navigation and acquisition of predictions by researchers and drug developers. The intuitive design of the application streamlines the prediction process and ensures efficient use of the software.

💾 Export and Save Results: Users are afforded the opportunity to export predicted pIC50 values and associated data for further analysis or integration into their research workflow. Additionally, the application enables users to save and organize prediction results for future reference.

Although the application provides state-of-the-art machine learning models that have been meticulously trained to provide accurate pIC50 predictions for human protein inhibitors, it is crucial to acknowledge that its outcomes may not align with reality. This is particularly evident in instances where there is no actual interaction between the substance and the ligand. Consequently, the pIC50 result obtained should be regarded with considerable caution and as a potential indicator of the substance's (rather than a definitive indicator of its) utility.

The article is currently in the process of being prepared for publication.