Biomedical Text Mining and Ontologies

This paper fine-tunes Llama 2 for rare disease concept normalization using HPO data, achieving high accuracy with synonym inclusion and typo handling.

This paper introduces AD-Mapper, a tool for semantic harmonization of Alzheimer's disease datasets, which outperforms a string-matching baseline.

This paper develops an annotation pipeline and two deep learning models for enzyme named entity recognition, achieving high performance with SciBERT-BiLSTM outperforming BioBERT-BiLSTM.

This paper investigates using Large Language Models for pathway curation, showing improved performance over existing methods with a GPT-3.5 fine-tuned model.

This scoping review summarizes the use of large language models in biomedical NLP tasks, highlighting their growing application in text classification, relation extraction, and named entity recognition.

The paper develops a web portal and NLP pipeline to extract and visualize gene-specific Alzheimer's disease publications from PubMed.

The paper maps PanglaoDB cell-type markers to Wikidata to enhance reusability and FAIRness, creating structured associations and demonstrating new biological insights.

The paper uses ontologies to integrate clinical, pathological, and molecular kidney data for precision medicine analysis.

This study evaluates GPT-3.5 and GPT-4 for identifying biological processes and regulatory relationships in wound healing texts, finding GPT-4 performs better without fine-tuning.

GO2Sum generates human-readable protein function summaries from GO terms using a fine-tuned T5 model, outperforming the base T5 model in creating function descriptions for UniProt entries.

The paper presents a human-in-the-loop system for identifying residue-level functional annotations from literature, achieving high precision, recall, and F1-measure.

The paper describes a structured curation process for the Gene2Phenotype project to systematically evaluate and organize genomic disease-gene relationships.

preon normalizes drug names and cancer types in large-scale data for precision oncology.

The paper fine-tunes BERT models to extract transcriptional regulatory interactions from bacterial literature, achieving high performance and reconstructing a significant portion of a bacterial transcriptional regulatory network.

The paper develops HSCLO38, a framework for connecting genomic features in a knowledge graph, and evaluates its performance with large-scale genomic data.

PuMA extracts and visualizes gene-celltype relations from PubMed to aid in celltype annotation.

Phenomics Assistant is a chat-based interface that uses LLMs to query the Monarch knowledge graph, improving reliability for non-expert users to explore disease-gene-phenotype relationships.

reguloGPT constructs a knowledge graph of molecular regulatory pathways using GPT-4 for entity recognition and relationship extraction, improving over existing methods and demonstrating utility in cancer research.

This paper evaluates 21 large language models for mining gene interactions and pathway knowledge, finding significant performance differences between API-based and open-source models.

The paper introduces CoNECo, a corpus for NER and NEN of protein complexes, achieving F1-scores of 73.7% and 61.2% with transformer-based and dictionary-based taggers, respectively.

The paper develops an ensemble transformer model using gene and chemical descriptions to extract drug-target interactions from biomedical literature, achieving high performance on a benchmark dataset.

The paper integrates the Medpipe microservice with Pannotator to enhance genomic annotation with immunological and subcellular location features.

RegulaTome provides a comprehensive corpus of biomedical relations, enabling improved relation extraction with a Transformer model achieving 66.6% F1-score.

The paper extends the Protein Kinase Ontology (ProKinO) to uncover new insights about dark kinases, identifying PAK5 as a frequently mutated kinase in cancers and highlighting its potential in drug discovery.

The paper introduces simona, an R package for semantic similarity analysis on bio-ontologies, benchmarking methods and showing topology-based approaches are effective for less-studied organisms.

The paper develops CeN-CAM, a semantic model based on GO-CAM, to represent causal relationships in C. elegans neural circuits and behaviors using literature data.

The paper introduces BioWiC, a benchmark to evaluate how well language models represent biomedical concepts in context, and establishes baselines using various large language models.

SKiM discovers and annotates biomedical knowledge using co-occurrence and transformer models to find A-B-C relationships between user-defined concepts.

The paper introduces a human-annotated full-text corpus for biomedical entities, including genes, proteins, diseases, and organisms, to support deep learning approaches in NLP.

PhenoID is a transformer-based NLP system that normalizes dysmorphology exam findings into HPO terms, achieving state-of-the-art performance on phenotype extraction from clinical notes.