Abstract Motivation Manual curation of biomedical literature is slow and error-prone and while large language models trained on general texts have shown to be useful for text summarisation, these methods lack the domain-specific expertise required to perform this task accurately. Here we describe the creation of the first microbiome-specific text corpus, use this to train deep learning algorithms for named-entity recognition (NER) and entity linking (EL), and demonstrate their use to meta-analyse microbiome literature. Results The training and validation set (n = 1410) contained a total of 90 150 annotations (both long form and abbreviations). Using the gold-standard test set (n = 288), with an inter-annotator agreement rate of 99.52% for NER and 88.31% for EL, the trained models were evaluated and our fine-tuned BioBERT model achieved an F1-score of 96% for NER surpassing a rule- and dictionary-based annotation pipeline (94%). For EL the accuracy obtained by the deep learning models greatly surpassed that of the pipeline (91% vs 69%). Evaluated across the entire available literature (n = 6927) across 14 domains, our models annotate an entire full-text document in only 7 seconds. Availability All codes are available for automatic annotation and model training, with instructions on how to deploy the model on new text, from GitHub at https://github.com/omicsNLP/microbELP and Zenodo at https://dx.doi.org/10.5281/zenodo.20613467. The redistributable, annotated training set and unannotated test set are made available from Zenodo at https://dx.doi.org/10.5281/zenodo.17305410 with the redistributable, human-labelled test set hosted as benchmark on Codabench at https://www.codabench.org/competitions/10913/ (for NER only) and at https://www.codabench.org/competitions/11581/ (for NER+EL) for evaluation. The annotated documents for all available literature are hosted separately on Zenodo at https://dx.doi.org/10.5281/zenodo.17288826.

MOTIVATION: Copy number variations (CNVs) are a major source of genomic variability and are especially significant in cancer. Until recently microarray technologies have been used to characterize CNVs in genomes. However, advances in next-generation sequencing technology offer significant opportunities to deduce copy number directly from genome sequencing data. Unfortunately cancer genomes differ from normal genomes in several aspects that make them far less amenable to copy number detection. For example, cancer genomes are often aneuploid and an admixture of diploid/non-tumor cell fractions. Also patient-derived xenograft models can be laden with mouse contamination that strongly affects accurate assignment of copy number. Hence, there is a need to develop analytical tools that can take into account cancer-specific parameters for detecting CNVs directly from genome sequencing data. RESULTS: We have developed WaveCNV, a software package to identify copy number alterations by detecting breakpoints of CNVs using translation-invariant discrete wavelet transforms and assign digitized copy numbers to each event using next-generation sequencing data. We also assign alleles specifying the chromosomal ratio following duplication/loss. We verified copy number calls using both microarray (correlation coefficient 0.97) and quantitative polymerase chain reaction (correlation coefficient 0.94) and found them to be highly concordant. We demonstrate its utility in pancreatic primary and xenograft sequencing data. AVAILABILITY AND IMPLEMENTATION: Source code and executables are available at https://github.com/WaveCNV. The segmentation algorithm is implemented in MATLAB, and copy number assignment is implemented Perl. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Large-scale international projects are underway to generate collections of knockout mouse mutants and subsequently to perform high throughput phenotype assessments, raising new challenges for computational researchers due to the complexity and scale of the phenotype data. Phenotypes can be described using ontologies in two differing methodologies. Traditionally an individual phenotypic character has either been defined using a single compound term, originating from a species-specific dedicated phenotype ontology, or alternatively by a combinatorial annotation, using concepts from a range of disparate ontologies, to define a phenotypic character as an entity with an associated quality (EQ). Both methods have their merits, which include the dedicated approach allowing use of community standard terminology, and the combinatorial approach facilitating cross-species phenotypic statement comparisons. Previously databases have favoured one approach over another. The EUMODIC project will generate large amounts of mouse phenotype data, generated as a result of the execution of a set of Standard Operating Procedures (SOPs) and will implement both ontological approaches to capture the phenotype data generated. RESULTS: For all SOPs a four-tier annotation is made: a high-level description of the SOP, to broadly define the type of data generated by the SOP; individual parameter annotation using the EQ model; annotation of the qualitative data generated for each mouse; and the annotation of mutant lines after statistical analysis. The qualitative assessments of phenodeviance are made at the point of data entry, using child PATO qualities to the parameter quality. To facilitate data querying by scientists more familiar with single compound terms to describe phenotypes, the mappings between the Mammalian Phenotype (MP) ontology and the EQ PATO model are exploited to allow querying via MP terms. CONCLUSION: Well-annotated and comparable phenotype databases can be achieved through the use of ontologically derived comparable phenotypic statements and have been implemented here by means of OBO compatible EQ annotations. The implementation we describe also sees scientists working seamlessly with ontologies through the assessment of qualitative phenotypes in terms of PATO qualities and the ability to query the database using community-accepted compound MP terms. This work represents the first time the combinatorial and single-dedicated approaches have both been implemented to annotate a phenotypic dataset.