The study of how the observable features of organisms, i.e., their phenotypes, result from the complex interplay between genetics, development, and the environment, is central to research in biology. Differences in the way that phenotypes of biodiverse species are described in text, however, has meant that it is difficult to use computer-based methods for analysis. Further, it has meant that researchers from different areas such as genetics cannot readily access or use these data in combination with their own. The overall aim of the Phenoscape initiative (www.phenoscape.org) is to develop the capacity for large-scale computational analysis of those phenotypes that distinguish species in nature. The focus, and intellectual merit, of this specific project has been to develop the capacity for large-scale analysis of phenotypes using a particular test case, the fin-to-limb transition, a time of major structural change as the first terrestrial vertebrates evolved from their aquatic ancestors. This was chosen owing to the large literature on both the phenotypic changes observed in the fossil record and on genes that may have been involved in those changes.

This project developed broadly applicable methods to mark-up text-based phenotypes with ontologies (standardized terms with built-in logic) so that they could be computed upon and then analyzed for their similarities and differences. The research increased the speed and scalability of computation resulting in rigorous, and user-friendly software that uses a set of phenotypes for a particular organism or genotype and retrieves statistically similar sets of phenotypes. The software integrates the logical structure of the ontologies along with statements of homology, phenotypes of biodiverse organisms, and phenotypes and associated genes of biomedical models in the Phenoscape Knowledgebase. This knowledgebase integrates data on over 5,000 vertebrate taxa and 20,000 evolutionary characters with hundreds of thousands of phenotypes from genes and gene expression from three vertebrate model organisms (zebrafish (Danio rerio), frog (Xenopus laevis), and mouse (Mus musculus)) and human. Experiments were conducted to determine how best to combine information about homology, or the descent of anatomical structures from a common ancestor, with ontology-annotated phenotype data. Software was developed to link the phenotypes from different species of vertebrate animals to the phenotypes and genes from mutant model organisms and vice versa. A user of the knowledgebase can thus compute the genes that might be involved in the evolutionary changes that produced the differences between species. As a capstone, an experiment was conducted to retrieve candidate genes from the knowledgebase for the vertebrate fin-to-limb transition based on the phenotypes of the relevant fossil taxa. The system successfully recovered most of the set of candidate genes proposed in the literature, while at the same time suggesting refinements to the expert predictions. Because phenotype text mark-up is time-consuming for humans and thus a major limitation in scaling up the approach, the project developed the first expert-curated dataset to be used to as a benchmark or ‘gold standard’ in assessing the efficacy of machine-based phenotypes. Project assessments using the benchmark point toward ways to better design software to assist human curators, and the use of the gold standard phenotype set will allow training and assessment of new tools to improve phenotype mark-up accuracy at scale. Finally, the project developed tools to automatically combine independently published data sets into large matrices of presence/absence phenotypes for taxa. The tools used the logic of ontologies to infer the presence or absence of a phenotype, and thus made available significant new data for biological research.

This research is one of the first to demonstrate the feasibility and utility of ontologies for data integration and knowledge discovery in evolutionary biology as well as the limitations and needs. The knowledgebase, ontologies, and over 30 publications with associated datasets and software are openly available for reuse by the scientific community. Contributions made by this project to community anatomy, quality, taxonomy, evidence ontologies and related resources are being leveraged by the broader biological and biomedical communities. Additional broader impacts from this project include training of undergraduate students, graduate students and postdocs at multiple U.S. institutions in the areas of biocuration, software development, semantic methods and support for programmers and curators at several institutions. Broader impacts also include several workshops and symposia on these emerging techniques for a research audience. The project developed partnerships with over 15 other organizations and institutions.

Last Modified: 09/13/2018
Modified by: Paula M Mabee