Welcome to ProBufEx (Database for Protein Expression, Purification and Characterization)

This website allows you to search for related buffers for protein purification and characterization using keywords or sentences, view related figures, and learn more about the database features. In addition, you can search relevant recombinant protein expression conditions using keywords. Please note that this database was constructed using an LLM-based pipeline, and its estimated accuracy is shown below.

Functions:

• Buffer Usage: Enter keywords, sentences, or a UniProt ID to find relevant results, including references, buffer details and functions, protein names, and DOIs. Each buffer entry is generated by An Advanced LLM-Powered Article Extraction Pipeline, which identifies buffer descriptions from protein-structure articles, links them to the relevant protein and experimental context, and summarizes their functional role. Natural-language queries are converted into embeddings and compared with the refined buffer index, allowing the search system to retrieve entries that are semantically related to the user's question rather than limited to exact keyword matches.
• Buffer Components: Within our database, proteins were systematically annotated and grouped using 1,145 unique UniProtKB keywords, which are classified into seven categories: biological process, cellular component, coding sequence diversity, domain, ligand, molecular function, and post-translational modification. You can input a keyword to retrieve and display statistical data on buffer components.
• Protein Expression: Supports keyword-based queries using UniProtKB annotations and retrieves recombinant protein production data from published literature.

How to Use:

Navigate using the menu at the top. For searches, enter a term in the search box and submit.

Overview of database statistics:

(A) Literature coverage and extraction success for PDB-associated articles. The ProBufEx pipeline processed 68,248 articles linked to direct depositions of experimentally determined protein 3D structures in the RCSB PDB, as well as 32,105 associated key citation articles. The panel summarizes extraction success among articles linked to direct depositions of experimentally determined protein 3D structures in the PDB.

(B) Manual evaluation of extraction accuracy for protein-buffer and expression-condition fields. The extraction of the protein purification and characterization buffers was evaluated on 1,218 eligible buffer records from 150 randomly sampled articles from our database. Expression-condition extraction was evaluated on 271 extracted culture conditions from a separate 150-article set. Induction strategy was evaluated on the eligible subset containing induction-strategy information (n = 161) associated with the E. coli expression system.

(C) Top 20 most frequent buffer categories identified in the database. Categories were assigned by a large language model (Mistral-Small-3.1-24B-Instruct).

(D) Distribution of expression systems used for recombinant protein production, showing both case counts and relative percentages.