Molecular Resistance Pathway Prediction System using SMILES-based Structural Features for Drug Response and Bioactivity Analysis

This paper introduces a compact deployable Streamlit-based system molecular property prediction with resistance pathway classification that shows the ability to use the minimal, well-defined features for the rapid prototyping in the drug discovery workflow. The application has two interactive streams, one is the estimator of the molecular weight with two explicit inputs i.e. number of hydrogen bond donors and number of hydrogen bond acceptors, and it performs the inference using a pre-trained model loaded from molecularmodel.pkl The second stream predicts resistance pathways based on accepting SMILES string (or csv file with a SMILES column) and obtaining a simple feature (SMILES length) and using a binary classifier loaded from resistance model. pkl to assign Resistant or Non-Resistant labels for the entries, with batch predictions being supported and make predictions in a tabular view. The codebase uses pandas for handling the data and joblib to serialize the model, as needed to process HDF5 files, and according to the project environment, scikit-learn and rdkit are among the project's dependencies, suggesting that the project supports traditional chemoinformatics tools. We describe the input engineering, outflows of inference, user interaction design and evaluate ready outputs implied by the code and emphasize the desire for reproducibility, for transparency of feature choices and light weight versatility for deployment path that can be aggressive for exemplary analyses and integration into larger pipelines.