Biology subjects

Posma, J. M.

Publications and source records attributed to Posma, J. M..

10 recordsLinked to original sources

bioRxiv2026

Relation Extraction for Diet, Non-Communicable Disease and Biomarker Associations (RECoDe): A CoDiet study

Diet plays a critical role in human health, with growing evidence linking dietary habits to disease outcomes. However, extracting structured dietary knowledge from biomedical literature remains challenging due to the lack of dedicated relation extraction datasets. To address this gap, we introduce RECoDe, a novel relation extraction (RE) dataset designed specifically for diet, disease, and related biomedical entities. RECoDe captures a diverse set of relation types, including a broad spectrum of positive association patterns and explicit negative examples, with over 5,000 human-annotated instances validated by up to five independent annotators. Furthermore, we benchmark various natural language processing (NLP) RE models, including BERT-based architectures and enhanced prompting techniques with locally deployed large language models (LLMs) to improve classification performance on underrepresented relation types. The best performing model gpt-oss-20B, a local LLM, achieved an F1-score of 64% for multi-class classification and 92% for binary classification using a hierarchical prompting strategy with a separate reflection step built in. To demonstrate the practical utility of RECoDe, we introduce the Contextual Co-occurrence Summarisation (Co-CoS) framework, which aggregates sentence-level relation extractions into document-level summaries and further integrates evidence across multiple documents. Co-CoS produces effect estimates consistent with established dietary knowledge, demonstrating its validity as a general framework for systematic evidence synthesis. AvailabilityThe code, models, and data will be made freely available upon acceptance.

bioinformatics↗

bioRxiv2025

A META-ANALYSIS OF GUT MICROBIOME RESEARCH IN MALNOURISHED AFRICAN POPULATIONS: A NATURAL LANGUAGE PROCESSING APPROACH

BackgroundMalnutrition still affects millions of children in Africa. Changes in the gut microbiome have been implicated in malnutrition, but there has been inconsistent nomenclature of microbes. This meta-analysis reviews the microbiome literature using natural language processing (NLP) methods. MethodologyWe searched PubMed for gut microbiome studies of undernourished children living in low-middle-income countries (LMICs). The primary analysis focused on continental coverage and study characteristics of microbiome research in sub-Saharan Africa. We also employed an NLP tool for normalising primary data from full-text publications in ss-Africa compared to other LMICs, and between diseased and healthy children. ResultsWe identified 16 studies. Most studies were conducted in Malawi and characterised the faecal microbiome using 16S rRNA sequencing. For comparison, 18 studies conducted in Bangladesh, India, Pakistan and Peru were included. With this, we identified frequently reported microbes that were distinctly identified in sub-Saharan Africa and highlighted possible signatures of an undernourished faecal microbiome across the globe. ConclusionThe consistent associations between elevated Pseudomonadota levels and severe acute malnutrition provides new insights into host-microbiome interactions in African contexts. However, the overlap between taxa associated with healthy and stunting underscores the need for further research to better inform potential targeted interventions in Africa.

genomics↗

bioRxiv2025

High-precision Biomedical Text Corpora for Multi-Entity Recognition: A CoDiet study

We present here five biomedical, multi-entity corpora that can be used as benchmarks for named-entity recognition (NER), targeted to literature on metabolic syndrome. The CoDiet-Gold corpus contains annotations for 500 full-text publications and 348,406 annotations. It is divided into CoDiet-Gold-public (450 documents) and CoDiet-Gold-private (50 documents). Each document was independently annotated by two human experts, with disagreements fully adjudicated by a third expert. The CoDiet-Electrum corpus (2,998,273 annotations) contains 4,423 publications that were annotated using case-insensitive matching of the surface forms with punctuation ignored, found in CoDiet-Gold-public. Finally, for the same 4,423 documents, two fully machine annotated corpora CoDiet-Bronze (2,938,738 annotations) and CoDiet-Silver (2,298,988 annotations), were created by utilising existing NER algorithms to annotate these. These corpora contain categories (organisms, disease, genes, proteins, metabolites) that add depth to existing corpora, as well as new categories that do not appear in other corpora (food, dietary methods, sample types, computational methods, study methodology, population characteristics, data types, and microbiome).

bioinformatics↗

bioRxiv2025

Microbial Named Entity Recognition and Normalisation for AI-assisted Literature Review and Meta-Analysis

MotivationManual curation of biomedical literature is slow and error-prone and while large language models (LLMs) 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 normalisation (NEN), and demonstrate their use to meta-analyse microbiome literature. MethodsWe developed an automated pipeline to annotate all mentions of bacteria, archaea, and fungi in 1,410 full-text microbiome articles. We manually annotated (gold-standard) a separate test set of 288 documents. We trained different transformer-based language models for microbiome recognition and normalisation to taxonomic identifiers and evaluate their performance using the precision, recall, F1-score, and accuracy on the test set. The best models were used to automatically annotate all available Open Access, full-text microbiome articles (n=6,927) and identify taxa that are significantly overrepresented across 14 domains. ResultsThe training and validation set contained a total of 90,150 annotations (both long form and abbreviations). Using the gold-standard test set, with an inter-annotator agreement rate of 99.52% for NER and 88.31% for NEN, 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 NEN the accuracy obtained by the deep learning models greatly surpassed that of the pipeline (91% vs 69%). Evaluated across the entire available literature, our models annotate an entire full-text document in only 7 seconds. ConclusionOur algorithms have near perfect precision and greatly speed up the process of annotating microbes in full-text articles. We demonstrated the capabilities of these methods by analysing the entire available literature and describe the taxa associated with each of the domains in our meta-analysis, and exemplify how these methods can be integrated into literature review workflows improving both the speed and accuracy of results. AvailabilityAll codes and data for automatic annotation, model training, and generation of taxonomic trees visualising the data will be made available following peer review with instructions on how to deploy the model on new texts from https://github.com/omicsNLP/microbELP.

bioinformatics↗

bioRxiv2024

Topological embedding and directional feature importance in ensemble classifiers for multi-class classification

Cancer is the second leading cause of disease-related death worldwide, and machine learning-based identification of novel biomarkers is crucial for improving early detection and treatment of various cancers. A key challenge in applying machine learning to high-dimensional data is deriving important features in an interpretable manner to provide meaningful insights into the underlying biological mechanisms We developed a class-based directional feature importance (CLIFI) metric for decision tree methods and demonstrated its use for the The Cancer Genome Atlas proteomics data. We incorporated this metric into four algorithms, Random Forest (RF), LAtent VAriable Stochastic Ensemble of Trees (LAVASET), and Gradient Boosted Decision Trees (GBDTs), and a new extension incorporating the LAVA step into GBDTs (LAVABOOST). Both LAVA methods incorporate topological information from protein interactions into the decision function. The different models performance in classifying 28 cancers resulted in F1-scores of 93% (RF), 92% (LAVASET), 89% (LAVABOOST) and 86% (GBDT), with no method outperforming all others for individual cancer type prediction. The CLIFI metric allowed the visualisation of the model decision making functions, and the distributions indicated heterogeneity in several proteins (MYH11, ER, BCL2) for different cancer types (including brain glioma, breast, kidney, thyroid and prostate cancer). We have developed an integrated, directional feature importance metric for multi-class decision tree-based classification models that facilitates interpretable feature importance assessment. The CLIFI metric can be used in conjunction with incorporating topological information into the decision functions of models to add inductive bias for improved interpretability. AvailabilityAll codes are available for data curation from https://github.com/EloisaRL/TCGA-proteomics-pipeline and the LAVASET (v1.0) package from https://github.com/melkasapi/LAVASET.

bioinformatics↗

bioRxiv2023

LAVASET: Latent Variable Stochastic Ensemble of Trees. A novel ensemble method for correlated datasets.

MotivationRandom Forests (RFs) can deal with a large number of variables, achieve reasonable prediction scores, and yield highly interpretable feature importance values. As such, RFs are appropriate models for feature selection and further dimension reduction (DR). However, RFs are often not appropriate for correlated datasets due to their mode of selecting individual features for splitting. Addressing correlation relationships in high dimensional datasets is imperative for reducing the number of variables that are assigned high importance, hence making the DR most efficient. Here, we propose the LAtent VAriable Stochastic Ensemble of Trees (LAVASET) method that derives latent variables based on the distance characteristics of each feature and aims to incorporate the correlation factor in the splitting step. ResultsWithout compromising on performance in the majority of examples, LAVASET outperforms RF by accurately determining feature importance across all correlated variables and ensuring proper distribution of importance values. LAVASET yields mostly non-inferior prediction accuracies to traditional RFs when tested in simulated and real 1D datasets, as well as more complex and high-dimensional 3D datatypes. Unlike traditional RFs, LAVASET is unaffected by single important noisy features (false positives), as it considers the local neighbourhood. LAVASET, therefore, highlights neighbourhoods of features, reflecting real signals that collectively impact the models predictive ability. LAVASET is freely available as a standalone package from https://github.com/melkasapi/LAVASET.

bioinformatics↗

bioRxiv2023

Vocabulary Matters: An Annotation Pipeline and Two Deep Learning Algorithms for Enzyme Named Entity Recognition

Enzymes are indispensable substances in many biological processes. With biomedical literature growing exponentially, it becomes more difficult to review the literature effectively. Hence, text-mining techniques are needed to facilitate and speed up literature review. The aims of this study are to create a corpus with annotated enzymes to train and evaluate enzyme named-entity recognition (NER) models. A novel pipeline was built using a combination of dictionary matching and rulebased keyword searching to automatically annotate enzyme entities in over 4,800 biomedical full texts. Two Bidirectional Long Short-Term Memory (BiLSTM) networks using BioBERT and SciBERT as tokeniser and word embedding layers were trained on this corpus and evaluated on a manually annotated test set of 526 full-text publications. The dictionary- and rule-based annotation pipeline achieved an F1-score of 0.863 (precision 0.996, recall 0.762). The SciBERT-BiLSTM model (F1-score 0.965, precision 0.981, recall 0.954) largely out-performed the BioBERT-BiLSTM model (F1-score 0.955, precision 0.981, recall 0.937). This study contributed a novel dictionary- and rule-based automatic pipeline with almost perfect precision which runs in a matter of seconds on a standard laptop. Both deep learning (DL) models achieved state-of-the-art performance (F1>0.95) for enzyme NER, with the SciBERT-based model outperforming the BioBERT-based model in terms of recall, demonstrating the vocabulary used by models matters. The proposed pipeline with the DL models can facilitate more effective enzyme text-mining and information extraction research for literature review and are the first algorithms specifically for enzyme NER. AvailabilityAll codes are available for automatic annotation and model training (including data), with instructions on how to deploy the model on new text, from https://github.com/omicsNLP/enzymeNER.

bioinformatics↗

bioRxiv2022

MetaboListem and TABoLiSTM: Two Deep Learning Algorithms for Metabolite Named Entity Recognition

Reviewing the metabolomics literature is becoming increasingly difficult because of the rapid expansion of relevant journal literature. Text-mining technologies are therefore needed to facilitate more efficient literature review. Here we contribute a standardised corpus of full-text publications from metabolomics studies and describe the development of two new metabolite named entity recognition (NER) methods. We introduce two deep learning methods for metabolite NER based on Bidirectional Long Short-Term Memory (BiLSTM) networks incorporating different transfer learning techniques. Our first model (MetaboListem) follows prior methodology using GloVe word embeddings. Our second model exploits BERT and BioBERT for embedding and is named TABoLiSTM (Transformer-Affixed BiLSTM). The methods are trained on a novel corpus annotated using rule-based methods, and evaluated on manually annotated metabolomics articles. MetaboListem (F1 score 0.890, precision 0.892, recall 0.888) and TABoLiSTM (BioBERT version: F1 score 0.909, precision 0.926, recall 0.893) have achieved state-of-the-art performance on metabolite NER. A corpus with >1,200 full-text Open Access metabolomics publications and >116,000 annotated metabolites was created. This work demonstrates that deep learning algorithms are capable of identifying metabolite names accurately and efficiently in text. The proposed corpus and NER algorithms can be used for metabolomics text-mining tasks such as information retrieval, document classification and literature-based discovery. AvailabilityThe corpus and NER algorithms are freely available with detailed instructions from Github at https://github.com/omicsNLP/MetaboliteNER.

bioinformatics↗

bioRxiv2021

Auto-CORPus: Automated and Consistent Outputs from Research Publications

To analyse large corpora using machine learning and other Natural Language Processing (NLP) algorithms, the corpora need to be standardised. The BioC format is a community-driven simple data structure for sharing text and annotations, however there is limited access to biomedical literature in BioC format and a lack of bioinformatics tools to convert online publication HTML formats to BioC. We present Auto-CORPus (Automated pipeline for Consistent Outputs from Research Publications), a novel NLP tool for the standardisation and conversion of publication HTML and table image files to three convenient machine-interpretable outputs to support biomedical text analytics. Firstly, Auto-CORPus can be configured to convert HTML from various publication sources to BioC. To standardise the description of heterogenous publication sections, the Information Artifact Ontology is used to annotate each section within the BioC output. Secondly, Auto-CORPus transforms publication tables to a JSON format to store, exchange and annotate table data between text analytics systems. The BioC specification does not include a data structure for representing publication table data, so we present a JSON format for sharing table content and metadata. Inline tables within full-text HTML files and linked tables within separate HTML files are processed and converted to machine-interpretable table JSON format. Finally, Auto-CORPus extracts abbreviations declared within publication text and provides an abbreviations JSON output that relates an abbreviation with the full definition. This abbreviation collection supports text mining tasks such as named entity recognition by including abbreviations unique to individual publications that are not contained within standard bio-ontologies and dictionaries. AvailabilityThe Auto-CORPus package is freely available with detailed instructions from Github at https://github.com/omicsNLP/Auto-CORPus/.

bioinformatics↗

bioRxiv2019

A network-based data-mining approach to investigate indole-related microbiota-host co-metabolism

MotivationIndoles have been shown to play a significant role in cardiometabolic disorders. While some individual bacterial species are known to produce indole-adducts, to our best knowledge no studies have made use of publicly available genome data to identify prokaryotes, specifically those associated with the human gut microbiota, contributing to the indole metabolic network.\n\nResultsHere, we propose a computational strategy, comprising the integration of KEGG and BLAST, to identify prokaryote-specific metabolic reactions relevant for the production of indoles, as well as to predict new members of the human gut microbiota potentially involved in these reactions. By identifying relevant prokaryotic species for further validation studies in vitro, this strategy represents a useful approach for those interrogating the metabolism of other gut-derived microbial metabolites relevant to human health.\n\nAvailabilityAll R scripts and files (gut microbial dataset, FASTA protein sequences, BLASTP output files) are available from https://github.com/AndreaRMICL/Microbial_networks.\n\nContactARM: andrea.rodriguez-martinez13@imperial.ac.uk; LH: lesley.hoyles@ntu.ac.uk.

bioinformatics↗