skip to main content


The NSF Public Access Repository (NSF-PAR) system and access will be unavailable from 5:00 PM ET until 11:00 PM ET on Friday, June 21 due to maintenance. We apologize for the inconvenience.

Title: Combining Machine Learning & Reasoning for Biodiversity Data Intelligence
The current crisis in global natural resource management makes it imperative that we better leverage the vast data sources associated with taxonomic entities (such as recognized species of plants and animals), which are known collectively as biodiversity data. However, these data pose considerable challenges for artificial intelligence: while growing rapidly in volume, they remain highly incomplete for many taxonomic groups, often show conflicting signals from different sources, and are multi-modal and therefore constantly changing in structure. In this paper, we motivate, describe, and present a novel workflow combining machine learning and automated reasoning, to discover patterns of taxonomic identity and change - i.e. “taxonomic intelligence” - leading to scalable and broadly impactful AI solutions within the bio-data realm.  more » « less
Award ID(s):
Author(s) / Creator(s):
; ; ; ;
Publisher / Repository:
Date Published:
Journal Name:
Proceedings of the AAAI Conference on Artificial Intelligence
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Making the most of biodiversity data requires linking observations of biological species from multiple sources both efficiently and accurately (Bisby 2000, Franz et al. 2016). Aggregating occurrence records using taxonomic names and synonyms is computationally efficient but known to experience significant limitations on accuracy when the assumption of one-to-one relationships between names and biological entities breaks down (Remsen 2016, Franz and Sterner 2018). Taxonomic treatments and checklists provide authoritative information about the correct usage of names for species, including operational representations of the meanings of those names in the form of range maps, reference genetic sequences, or diagnostic traits. They increasingly provide taxonomic intelligence in the form of precise description of the semantic relationships between different published names in the literature. Making this authoritative information Findable, Accessible, Interoperable, and Reusable (FAIR; Wilkinson et al. 2016) would be a transformative advance for biodiversity data sharing and help drive adoption and novel extensions of existing standards such as the Taxonomic Concept Schema and the OpenBiodiv Ontology (Kennedy et al. 2006, Senderov et al. 2018). We call for the greater, global Biodiversity Information Standards (TDWG) and taxonomy community to commit to extending and expanding on how FAIR applies to biodiversity data and include practical targets and criteria for the publication and digitization of taxonomic concept representations and alignments in taxonomic treatments, checklists, and backbones. As a motivating case, consider the abundantly sampled North American deer mouse— Peromyscus maniculatus (Wagner 1845)—which was recently split from one continental species into five more narrowly defined forms, so that the name P. maniculatus is now only applied east of the Mississippi River (Bradley et al. 2019, Greenbaum et al. 2019). That single change instantly rendered ambiguous ~7% of North American mammal records in the Global Biodiversity Information Facility (n=242,663, downloaded 2021-06-04; 2021) and ⅓ of all National Ecological Observatory Network (NEON) small mammal samples (n=10,256, downloaded 2021-06-27). While this type of ambiguity is common in name-based databases when species are split, the example of P. maniculatus is particularly striking for its impact upon biological questions ranging from hantavirus surveillance in North America to studies of climate change impacts upon rodent life-history traits. Of special relevance to NEON sampling is recent evidence suggesting deer mice potentially transmit SARS-CoV-2 (Griffin et al. 2021). Automating the updating of occurrence records in such cases and others will require operational representations of taxonomic concepts—e.g., range maps, reference sequences, and diagnostic traits—that are FAIR in addition to taxonomic concept alignment information (Franz and Peet 2009). Despite steady progress, it remains difficult to find, access, and reuse authoritative information about how to apply taxonomic names even when it is already digitized. It can also be difficult to tell without manual inspection whether similar types of concept representations derived from multiple sources, such as range maps or reference sequences selected from different research articles or checklists, are in fact interoperable for a particular application. The issue is therefore different from important ongoing efforts to digitize trait information in species circumscriptions, for example, and focuses on how already digitized knowledge can best be packaged to inform human experts and artifical intelligence applications (Sterner and Franz 2017). We therefore propose developing community guidelines and criteria for FAIR taxonomic concept representations as "semantic artefacts" of general relevance to linked open data and life sciences research (Le Franc et al. 2020). 
    more » « less
  2. We provide an overview and update on initiatives and approaches to add taxonomic data intelligence to distributed biodiversity knowledge networks. "Taxonomic intelligence" for biodiversity data is defined here as the ability to identify and renconcile source-contextualized taxonomic name-to-meaning relationships (Remsen 2016). We review the scientific opportunities, as well as information-technological and socio-economic pathways - both existing and envisioned - to embed de-centralized taxonomic data intelligence into the biodiversity data publication and knowledge intedgration processes. We predict that the success of this project will ultimately rest on our ability to up-value the roles and recognition of systematic expertise and experts in large, aggregated data environments. We will argue that these environments will need to adhere to criteria for responsible data science and interests of coherent communities of practice (Wenger 2000, Stoyanovich et al. 2017). This means allowing for fair, accountable, and transparent representation and propagation of evolving systematic knowledge and enduring or newly apparent conflict in systematic perspective (Sterner and Franz 2017, Franz and Sterner 2018, Sterner et al. 2019). We will demonstrate in principle and through concrete use cases, how to de-centralize systematic knowledge while maintaining alignments between congruent or concflicting taxonomic concept labels (Franz et al. 2016a, Franz et al. 2016b, Franz et al. 2019). The suggested approach uses custom-configured logic representation and reasoning methods, based on the Region Connection Calculus (RCC-5) alignment language. The approach offers syntactic consistency and semantic applicability or scalability across a wide range of biodiversity data products, ranging from occurrence records to phylogenomic trees. We will also illustrate how this kind of taxonomic data intelligence can be captured and propagated through existing or envisioned metadata conventions and standards (e.g., Senderov et al. 2018). Having established an intellectual opportunity, as well as a technical solution pathway, we turn to the issue of developing an implementation and adoption strategy. Which biodiversity data environments are currently the most taxonomically intelligent, and why? How is this level of taxonomic data intelligence created, maintained, and propagated outward? How are taxonomic data intelligence services motivated or incentivized, both at the level of individuals and organizations? Which "concerned entities" within the greater biodiversity data publication enterprise are best positioned to promote such services? Are the most valuable lessons for biodiversity data science "hidden" in successful social media applications? What are good, feasible, incremental steps towards improving taxonomic data intelligence for a diversity of data publishers? 
    more » « less
  3. null (Ed.)
    “What is crucial for your ability to communicate with me… pivots on the recipient’s capacity to interpret—to make good inferential sense of the meanings that the declarer is able to send” (Rescher 2000, p148). Conventional approaches to reconciling taxonomic information in biodiversity databases have been based on string matching for unique taxonomic name combinations (Kindt 2020, Norman et al. 2020). However, in their original context, these names pertain to specific usages or taxonomic concepts, which can subsequently vary for the same name as applied by different authors. Name-based synonym matching is a helpful first step (Guala 2016, Correia et al. 2018), but may still leave considerable ambiguity regarding proper usage (Fig. 1). Therefore, developing "taxonomic intelligence" is the bioinformatic challenge to adequately represent, and subsequently propagate, this complex name/usage interaction across trusted biodiversity data networks. How do we ensure that senders and recipients of biodiversity data not only can share messages but do so with “good inferential sense” of their respective meanings? Key obstacles have involved dealing with the complexity of taxonomic name/usage modifications through time, both in terms of accounting for and digitally representing the long histories of taxonomic change in most lineages. An important critique of proposals to use name-to-usage relationships for data aggregation has been the difficulty of scaling them up to reach comprehensive coverage, in contrast to name-based global taxonomic hierarchies (Bisby 2011). The Linnaean system of nomenclature has some unfortunate design limitations in this regard, in that taxonomic names are not unique identifiers, their meanings may change over time, and the names as a string of characters do not encode their proper usage, i.e., the name “Genus species” does not specify a source defining how to use the name correctly (Remsen 2016, Sterner and Franz 2017). In practice, many people provide taxonomic names in their datasets or publications but not a source specifying a usage. The information needed to map the relationships between names and usages in taxonomic monographs or revisions is typically not presented it in a machine-readable format. New approaches are making progress on these obstacles. Theoretical advances in the representation of taxonomic intelligence have made it increasingly possible to implement efficient querying and reasoning methods on name-usage relationships (Chen et al. 2014, Chawuthai et al. 2016, Franz et al. 2015). Perhaps most importantly, growing efforts to produce name-usage mappings on a medium scale by data providers and taxonomic authorities suggest an all-or-nothing approach is not required. Multiple high-profile biodiversity databases have implemented internal tools for explicitly tracking conflicting or dynamic taxonomic classifications, including eBird using concept relationships from AviBase (Lepage et al. 2014); NatureServe in its Biotics database; iNaturalist using its taxon framework (Loarie 2020); and the UNITE database for fungi (Nilsson et al. 2019). Other ongoing projects incorporating taxonomic intelligence include the Flora of Alaska (Flora of Alaska 2020), the Mammal Diversity Database (Mammal Diversity Database 2020) and PollardBase for butterfly population monitoring (Campbell et al. 2020). 
    more » « less
  4. Mosquito-borne diseases continue to ravage humankind with >700 million infections and nearly one million deaths every year. Yet only a small percentage of the >3500 mosquito species transmit diseases, necessitating both extensive surveillance and precise identification. Unfortunately, such efforts are costly, time-consuming, and require entomological expertise. As envisioned by the Global Mosquito Alert Consortium, citizen science can provide a scalable solution. However, disparate data standards across existing platforms have thus far precluded truly global integration. Here, utilizing Open Geospatial Consortium standards, we harmonized four data streams from three established mobile apps—Mosquito Alert, iNaturalist, and GLOBE Observer’s Mosquito Habitat Mapper and Land Cover—to facilitate interoperability and utility for researchers, mosquito control personnel, and policymakers. We also launched coordinated media campaigns that generated unprecedented numbers and types of observations, including successfully capturing the first images of targeted invasive and vector species. Additionally, we leveraged pooled image data to develop a toolset of artificial intelligence algorithms for future deployment in taxonomic and anatomical identification. Ultimately, by harnessing the combined powers of citizen science and artificial intelligence, we establish a next-generation surveillance framework to serve as a united front to combat the ongoing threat of mosquito-borne diseases worldwide. 
    more » « less
  5. Advances in genomics and transcriptomics accompanying the rapid accumulation of omics data have provided new tools that have transformed and expanded the traditional concepts of model fungi. Evolutionary genomics and transcriptomics have flourished with the use of classical and newer fungal models that facilitate the study of diverse topics encompassing fungal biology and development. Technological advances have also created the opportunity to obtain and mine large datasets. One such continuously growing dataset is that of the Sordariomycetes, which exhibit a richness of species, ecological diversity, economic importance, and a profound research history on amenable models. Currently, 3,574 species of this class have been sequenced, comprising nearly one-third of the available ascomycete genomes. Among these genomes, multiple representatives of the model genera Fusarium , Neurospora , and Trichoderma are present. In this review, we examine recently published studies and data on the Sordariomycetes that have contributed novel insights to the field of fungal evolution via integrative analyses of the genetic, pathogenic, and other biological characteristics of the fungi. Some of these studies applied ancestral state analysis of gene expression among divergent lineages to infer regulatory network models, identify key genetic elements in fungal sexual development, and investigate the regulation of conidial germination and secondary metabolism. Such multispecies investigations address challenges in the study of fungal evolutionary genomics derived from studies that are often based on limited model genomes and that primarily focus on the aspects of biology driven by knowledge drawn from a few model species. Rapidly accumulating information and expanding capabilities for systems biological analysis of Big Data are setting the stage for the expansion of the concept of model systems from unitary taxonomic species/genera to inclusive clusters of well-studied models that can facilitate both the in-depth study of specific lineages and also investigation of trait diversity across lineages. The Sordariomycetes class, in particular, offers abundant omics data and a large and active global research community. As such, the Sordariomycetes can form a core omics clade, providing a blueprint for the expansion of our knowledge of evolution at the genomic scale in the exciting era of Big Data and artificial intelligence, and serving as a reference for the future analysis of different taxonomic levels within the fungal kingdom. 
    more » « less