Biomedical Text Mining and Ontologies

This paper constructs a knowledge graph of CAR T cell signaling interactions using automated text mining and large language models to aid in CAR design.

PHENOCAUZ maps symptoms to molecular determinants using Mendelian disease biology, identifying candidate symptom drivers and therapeutic pathways.

The paper introduces a framework to generate synthetic phenotype data using HPO structure and trains a GNN to prioritize disease genes, outperforming existing methods on real datasets.

The paper introduces a transformer-based language model to identify computable phenotype definitions in biomedical literature and integrates it into a platform for literature curation.

The paper introduces Kg4j, a framework for building tailored knowledge graphs to explore disease-exposure relationships, applied to endometriosis and persistent pollutants.

This study analyzes 21 years of Gene Ontology (GO) releases to understand its structural and annotation changes over time, identifying a shift toward stability around 2017.

AOPGraphExplorer 2.0 provides an interactive platform for exploring and annotating adverse outcome pathways using a graph-based approach.

KG-Orchestra enriches biomedical knowledge graphs by combining manual curation and automated LLM methods to improve granularity and evidence quality.

The paper creates a manually annotated corpus of cell phenotype mentions and shows its use in improving named entity recognition and entity linking models.

The paper develops a comprehensive HPO profile for Fanconi anemia by integrating clinical guidelines and LLM-based ontology extraction, identifying significant gaps in existing annotations.

MetaKnogic-Alpha transforms biomedical literature into a hyper-relational knowledge base for metabolic reasoning, achieving high mechanistic accuracy through automated synthesis and biochemical grounding.

The paper develops TVBase, a platform that projects biomedical knowledge onto a 3D brain atlas to support neuroimaging analysis and hypothesis generation.

RAGulate improves the assessment of transcription factor-target relationships by integrating retrieval-augmented generation with a regulatory knowledge base to reduce hallucinations and enhance evidence faithfulness.

The paper introduces dgiLIT, an AI-driven method to prioritize and curate drug-gene interactions from biomedical literature for inclusion in DGIdb, achieving high validation rates.

EPITOME is a multimodal pipeline that combines OCR, text matching, and VLMs to automate and enhance epitope data curation from scientific papers.

DISGENET integrates genotype-phenotype data to support precision medicine and drug development through comprehensive, up-to-date associations and tools.

Alvessa improves accuracy in functional genomics and drug target assessment by integrating evidence-based reasoning and verification.

This paper defines a cell-cell relation extraction task, constructs annotated corpora, and evaluates models under realistic constraints, finding performance remains around 70% accuracy.

Specifind automates the extraction of species occurrence data from scientific literature by identifying species names, geographic references, and their relationships.

This paper applies large language models to assist in gene annotation into Reactome pathways, showing promising results in supporting curation workflows.

GENET streamlines hypothesis generation for gene-disease associations by integrating AI-driven network analysis, literature mining, clustering, and visualization.

React-to-Me, a retrieval-augmented chatbot grounding GPT-4o-mini in curated Reactome pathways with hyperlinked citations, earned higher blinded expert ratings than ungrounded GPT-4o-mini (odds ratio 2.01).

This study evaluates LLMs for microbial phenotype annotation, finding varied performance across models and traits, with confidence scores correlating with accuracy.

The paper develops an automated pipeline to map genetic developmental disorder literature to Gene2Phenotype using NLP models, achieving high accuracy and scalability for biocuration.

This paper introduces a multi-agent system using large language models to enhance gene set databases by incorporating biological context from PubMed abstracts.

The paper compares semantic similarity algorithms for differential diagnosis and clustering using HPO, finding Phenomizer performs best for ranking and non-normalized algorithms for clustering.

The paper develops a model to predict gene relevance in behavioral research using a large language model, revealing shared gene signatures between molecular and behavioral studies.

SOORENA identifies autoregulatory mechanisms in biological networks by analyzing PubMed abstracts, finding 97,657 protein-specific records and generating a curated database.

The paper introduces a tool that links protein residue mentions in text to their corresponding positions in protein structures, enabling integrated exploration with experimental evidence.

The paper fine-tunes a model to extract four research goal criteria from biomedical abstracts using the RG4C dataset, achieving an overall F-score of 0.59.