HMRPred: A Machine Learning-Based Web Resource for Identification of Heavy Metal Resistance Proteins

Heavy metal contamination poses a significant threat to environmental health, agriculture, and microbial ecosystems, necessitating the identification of molecular components that confer resistance. Heavy metal resistance (HMR) proteins enable organisms to survive toxic metal exposure through mechanisms such as efflux transport, enzymatic detoxification, and metal sequestration. However, the diversity and functional overlap of these proteins across taxa present challenges for reliable identification using conventional homology-based methods. Furthermore, current machine learning approaches for resistance gene prediction primarily focus on antibiotics, with no comprehensive resource available for systematically classifying HMR proteins across multiple metals and biological domains. To address this, we developed HMRPred, a machine learning-based predictive framework for the identification of HMR proteins across ten metals of concern: arsenic, cadmium, chromium, copper, iron, lead, mercury, nickel, silver, and zinc. Curated datasets comprising experimentally validated resistance and non-resistance proteins were used to extract a comprehensive set of sequence-derived features, including amino acid composition and physicochemical descriptors. For each metal, optimized classifiers were trained using various machine learning algorithms, achieving high performance with an AUC-ROC of more than 98% in both cross-validation and independent testing. HMRPred is deployed as a web-accessible resource (available at https://hmrpred.streamlit.app/), allowing researchers to submit protein sequences and obtain predictions with confidence scores. By facilitating genome-wide annotation of metal resistance determinants, HMRPred supports applications in bioremediation, environmental microbiology, phytoremediation, and synthetic biology.