LibFminer

LibFMiner implements a method for efficiently mining relevant tree-shaped molecular fragments, each representing a geometrical classes, with minimum frequency and statistical constraints. Experimental results with four test datasets suggest that the approach is able to optimize correlation with chemical endpoints as well as inter-feature entropy. The mined fragments represent classes of fragments sharing a common core structure (backbone). The proposed method reduces feature set sizes and runtime by >90% and >60% compared to ordinary tree mining. Validation with several publicly available carcinogenicity datasets shows that their expressiveness is comparable to the complete set of tree-shaped fragments and is significantly higher than linear fragments.

API documentation is available from http://www.maunz.de/libfminer-doc/. The documentation gives example programs, explains constructor usage and options.

The official FMiner application that uses this library is available from http://github.com/amaunz/fminer/tree/master, documentation is available from http://www.maunz.de/fminer-doc/.

Latest scientific documentation is available from http://github.com/amaunz/fminer-doc/tree/master.

• Homepage: http://www.maunz.de/libfminer-doc/
• Subproject of: OpenTox
• Data: No
• Software: Yes
• Programming language: C++, bindings can be automatically generated for ruby, Java, Python and many others
• Operating system: Linux, Windows planned
• Dependencies: OpenBabel, GSL
• Input format: SMILES strings, gSpan format
• Output format: SMARTS patterns
• Version control: git
• License: GNU GPL