Abstract
Motivation: The protein data bank (PDB) currently holds over 140,000 biomolecular structures and continues to release new structures on a weekly basis. The PDB is an essential resource to the structural bioinformatics community to develop software that use, categorize, and analyze such data. New computational biology methods are evaluated using custom benchmarking sets derived as subsets of 3D experimentally determined structures from the PDB. Currently, these benchmarking subsets are manually curated with custom scripts developed in a non-standardized manner that results in slow distribution and updates with new experimental structures. Finally, there are scarcity of tools available to rapidly query 3D descriptors of the entire PDB to derive subsequent benchmarking sets in a standard manner. Approach: Our solution is, Lemon, a C++11 framework that provides a methodology for selecting biomolecules based on precalculated and user derived 3D descriptors. This framework can parse and characterize the entire PDB in less than twenty minutes on modern, multithreaded hardware. The speed in parsing is obtained by using the recently developed MacroMolecule Transmission Format (MMTF) to reduce computational load of reading PDB files. We have incorporated C++ lambda functions that offer extensive flexibility to analyze and categorize the PDB by allowing the user to write custom functions to suite their objective. We think Lemon will become a one-stop-shop to quickly mine the entire PDB to generate custom benchmarking datasets for the entire scientific community. The Lemon software is available as a C++ header library along with example functions at https://github.com/chopralab/lemon.
