CEM 2.0: Co-evolution–guided mining of superior extremophilic proteins via a curated proteome database and protein language model

Extremophilic proteins offer immense potential for diverse applications; however, their discovery is hindered by the reliance on predefined templates and domain-specific knowledge. Therefore, new methods are required to mine extremophilic proteins without relying on existing templates, enabling a comprehensive exploration of functional extremophilic proteins across the entire protein universe. Here, we propose a co-evolution–guided protein mining framework, CEM 2.0 (Co-Evolution Mining 2.0), comprising three synergistic components: (1) an extensive and diverse extremophilic proteome database covering six extremophilic features, (2) a multidimensional evaluator for six extremophilic features, leveraging the protein language model ProteinBERT, and (3) an integrated and interactive mining pipeline equipped with advanced tools for computing more than 20 properties of proteins. This framework was implemented on an interactive web platform ( https://cem.sjtu.edu.cn/ ) equipped with intuitive visualization tools. We conducted wet-lab validation, performed a series of computational validations using published studies, and made recommendations for novel PETases, thereby demonstrating the effectiveness and potential of CEM 2.0 for the template-independent discovery of extremophilic proteins. CEM 2.0 is expected to mine functional extremophilic proteins across the protein universe in terms of evolutionary distance and serve as a valuable tool for protein engineering and synthetic biology.