MicroPIE

MicroPIE extracts phenotypic characters and character states from prokaryotic taxonomic descriptions to produce structured taxon-by-character matrices for downstream comparative and phylogenetic analyses.

Key Features:

• Natural language processing (NLP): Employs NLP techniques to parse and interpret complex taxonomic descriptions and identify candidate phenotypic information.
• Supervised classification (SVM): Uses a Support Vector Machine for supervised classification to identify phenotypic characters within sentences.
• Linguistic rules and term groups: Applies linguistic rules combined with known term groups to refine extraction and capture character states.
• Input and output formats: Processes clean taxonomic description text and outputs a taxon-by-character matrix with taxa as rows and 42 pre-defined characters (e.g., optimum growth temperature) as columns.
• Performance evaluation: Evaluated against a gold standard matrix and a matrix produced by undergraduate students, with 21 of 42 characters achieving Relaxed F1 > 0.80 and 16 of 42 characters scoring between 0.50 and 0.80.
• SVM impact on precision: Inclusion of the SVM-based character prediction component significantly increased precision relative to manual extraction methods.

Scientific Applications:

• Large-scale phenotypic data extraction: Enables conversion of legacy taxonomic descriptions into structured phenotypic data for prokaryotic organisms, including traits such as shape, metabolic substrates, and growth conditions.
• Input for phylogenetic analyses: Produces taxon-by-character matrices that can be used directly as input for phylogenetic and comparative analysis tools.
• Evolutionary and taxonomic studies: Facilitates study of the distribution and evolution of microbial traits in evolutionary biology, taxonomy, and microbiology.

Methodology:

Parses taxonomic descriptions with NLP, identifies potential phenotypic characters using a Support Vector Machine (SVM), and applies linguistic rules together with known term groups to extract characters and their states and produce a taxon-by-character matrix.