The Molecular Interaction (MINT) database is a structured repository of experimentally verified protein-protein interactions (PPIs), curated from peer-reviewed publications. MINT focuses on physical interactions between proteins, excluding genetic or computationally inferred ones. Over four years, MINT has undergone significant restructuring, leading to a more efficient database with richer annotations. As of September 2006, MINT contains over 95,000 curated physical interactions involving 27,461 proteins from 325 organisms. The database is accessible online through web interfaces and an FTP server, and includes HomoMINT, an inferred human protein interaction network based on orthologous interactions in model organisms. MINT uses the IntAct relational model and is based on the open-source PostgreSQL database. It allows data exploration through web-based interfaces, including query capabilities based on protein names, UniProt keywords, and identifiers from external databases. The database also supports sequence similarity searches and provides a graphical viewer for interaction networks. Each interaction is assigned a confidence level based on experimental methods and conditions, and all entries undergo quality control and curation according to the PSI-MI 2.5 standards. MINT is part of the International Molecular-Interaction Exchange (IMEx) consortium, which aims to avoid redundancy and share curation efforts. The database provides various data formats for export, including PSI1.0-XML, PSI2.5-XML, and flatfile. A web service allows computational access to interaction networks in different formats. MINT is supported by AIRC and the European Union FP6 projects, with funding for open access publication charges provided by FP6. The database is continuously updated and curated, with a focus on publications describing low-throughput experiments.The Molecular Interaction (MINT) database is a structured repository of experimentally verified protein-protein interactions (PPIs), curated from peer-reviewed publications. MINT focuses on physical interactions between proteins, excluding genetic or computationally inferred ones. Over four years, MINT has undergone significant restructuring, leading to a more efficient database with richer annotations. As of September 2006, MINT contains over 95,000 curated physical interactions involving 27,461 proteins from 325 organisms. The database is accessible online through web interfaces and an FTP server, and includes HomoMINT, an inferred human protein interaction network based on orthologous interactions in model organisms. MINT uses the IntAct relational model and is based on the open-source PostgreSQL database. It allows data exploration through web-based interfaces, including query capabilities based on protein names, UniProt keywords, and identifiers from external databases. The database also supports sequence similarity searches and provides a graphical viewer for interaction networks. Each interaction is assigned a confidence level based on experimental methods and conditions, and all entries undergo quality control and curation according to the PSI-MI 2.5 standards. MINT is part of the International Molecular-Interaction Exchange (IMEx) consortium, which aims to avoid redundancy and share curation efforts. The database provides various data formats for export, including PSI1.0-XML, PSI2.5-XML, and flatfile. A web service allows computational access to interaction networks in different formats. MINT is supported by AIRC and the European Union FP6 projects, with funding for open access publication charges provided by FP6. The database is continuously updated and curated, with a focus on publications describing low-throughput experiments.