MOTIVATION: Knowledge graphs (KGs) are a powerful approach for integrating heterogeneous data and making inferences in biology and many other domains, but a coherent solution for constructing, exchanging, and facilitating the downstream use of KGs is lacking. RESULTS: Here we present KG-Hub, a platform that enables standardized construction, exchange, and reuse of KGs. Features include a simple, modular extract-transform-load pattern for producing graphs compliant with Biolink Model (a high-level data model for standardizing biological data), easy integration of any OBO (Open Biological and Biomedical Ontologies) ontology, cached downloads of upstream data sources, versioned and automatically updated builds with stable URLs, web-browsable storage of KG artifacts on cloud infrastructure, and easy reuse of transformed subgraphs across projects. Current KG-Hub projects span use cases including COVID-19 research, drug repurposing, microbial-environmental interactions, and rare disease research. KG-Hub is equipped with tooling to easily analyze and manipulate KGs. KG-Hub is also tightly integrated with graph machine learning (ML) tools which allow automated graph ML, including node embeddings and training of models for link prediction and node classification. AVAILABILITY AND IMPLEMENTATION: https://kghub.org.

International Rice Research Institute, Philippines [email protected], WWW: www.irri.org

MOTIVATION: The high content of repetitive sequences in the genomes of many higher eukaryotes renders the task of annotating them computationally intensive. Presently, the only widely accepted method of searching and annotating transposable elements (TEs) in large genomic sequences is the use of the RepeatMasker program, which identifies new copies of TEs by pairwise sequence comparisons with a library of known TEs. Profile hidden Markov models (HMMs) have been used successfully in discovering distant homologs of known proteins in large protein databases, but this approach has only rarely been applied to known model TE families in genomic DNA. RESULTS: We used a combination of computational approaches to annotate the TEs in the finished genome of Oryza sativa ssp. japonica. In this paper, we discuss the strengths and the weaknesses of the annotation methods used. These approaches included: the default configuration of RepeatMasker using cross_match, an implementation of the Smith-Waterman-Gotoh algorithm; RepeatMasker using WU-BLAST for similarity searching; and the HMMER package, used to search for TEs with profile HMMs. All the results were converted into GFF format and post-processed using a set of Perl scripts. RepeatMasker was used in the case of most TE families. The WU-BLAST implementation of RepeatMasker was found to be manifold faster than cross_match with only a slight loss in sensitivity and was thus used to obtain the final set of data. HMMER was used in the annotation of the Mutator-like element (MULE) superfamily and the miniature inverted-repeat transposable element (MITE) polyphyletic group of families, for which large libraries of elements were available and which could be divided into well-defined families. The HMMER search algorithm was extremely slow for models over 1000 bp in length, so MULE families with members over 1000 bp long were processed with RepeatMasker instead. The main disadvantage of HMMER in this application is that, since it was developed with protein sequences in mind, it does not search the negative DNA strand. With the exception of TE families with essentially palindromic sequences, reverse complement models had to be created and run to compensate for this shortcoming. We conclude that a modification of RepeatMasker to incorporate libraries of profile HMMs in searches could improve the ability to detect degenerated copies of TEs. AVAILABILITY: The Perl scripts and TE sequences used in construction of the RepeatMasker library and the profile HMMs are available upon request.