GL4SDA: Predicting snoRNA-Disease Associations Using GNNs and LLM Embeddings

Computational and Structural Biotechnology Journal, Journal Year: 2025, Volume and Issue: 27, P. 1023 - 1033

Published: Jan. 1, 2025

Small nucleolar RNAs (snoRNAs) play essential roles in various cellular processes, and their associations with diseases are increasingly recognized. Identifying these snoRNA-disease relationships is critical for advancing our understanding of functional potential therapeutic implications. This work presents a novel approach, called GL4SDA, to predict using Graph Neural Networks (GNN) Large Language Models. Our methodology leverages the unique strengths heterogeneous graph structures model complex biological interactions. Differently from existing methods, we define set features able capture deeper information content related inner attributes both snoRNAs design GNN based on highly performing layers, which can maximize results this representation. We consider snoRNA secondary disease embeddings derived large language models obtain node features, respectively. By combining structural rich semantic diseases, construct feature-rich representation that improves predictive performance model. evaluate approach different architectures exploit capabilities many convolutional layers compare three other state-of-the-art graph-based predictors. GL4SDA demonstrates improved scores link prediction tasks its implication as tool exploring relationships. also validate findings through case studies about cancer highlighting practical application method real-world scenarios obtaining most important explainable artificial intelligence methods.

Language: Английский

AI-Powered Neurogenetics: Supporting Patient’s Evaluation with Chatbot

Genes, Journal Year: 2024, Volume and Issue: 16(1), P. 29 - 29

Published: Dec. 27, 2024

Artificial intelligence and large language models like ChatGPT Google's Gemini are promising tools with remarkable potential to assist healthcare professionals. This study explores Gemini's utility in assisting clinicians during the first evaluation of patients suspected neurogenetic disorders. By analyzing model's performance identifying relevant clinical features, suggesting differential diagnoses, providing insights into possible genetic testing, this research seeks determine whether these AI could serve as a valuable adjunct assessments. Ninety questions were posed (Versions 4o, 4, 3.5) Gemini: four about diagnosis, seven inheritance, estimable recurrence risks, available tests, patient management, each for six different rare disorders (Hereditary Spastic Paraplegia type 4 7, Huntington Disease, Fragile X-associated Tremor/Ataxia Syndrome, Becker Muscular Dystrophy, FacioScapuloHumeral Dystrophy). According results study, GPT chatbots demonstrated significantly better than Gemini. Nonetheless, all showed notable gaps diagnostic accuracy concerning level hallucinations. As expected, can empower assessing disorders, yet their effective use demands meticulous collaboration oversight from both neurologists geneticists.

Language: Английский