Genome-wide measurement of RNA dissociation from chromatin classifies transcripts by their dynamics and reveals rapid dissociation of enhancer lncRNAs

Ntini, Evgenia; Budach, Stefan; Vang Ørom, Ulf A.; Marsico, Annalisa

doi:10.1016/j.cels.2023.09.005

Cited by 1 publication

(2 citation statements)

References 87 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study, the authors reported 18 target gene-lncRNA regulation pairs. The second smaller dataset encompasses a set of enhancer-like lncRNAs and their potential cis targets determined via chromatin interactions, as defined in Ntini et al [107]. In a transductive setting, we evaluated the method by determining how many novel interactions from these datasets (i.e., those not included in the KG) could be identified at different probability thresholds, thereby constructing a recall curve for each dataset.…”

Section: Lncrna-target Prediction Taskmentioning

confidence: 99%

“…Further, we ranked the interaction partners of 42 lncRNAs, including PVT1, sourced from the studies of [107] and [106]. We considered a regulatory relationship between the lncRNA of interest and its target genes, when there existed any relation r for which the conditional probability exceeded a threshold t, p(v|u, q) ≥ t. Conversely, if no such relation exists, it is considered that there is no regulatory relationship.…”

Section: Lncrna-gene Target Prediction Taskmentioning

confidence: 99%

See 1 more Smart Citation

Path-based reasoning for biomedical knowledge graphs with BioPathNet

Hu,

Oleshko,

Firmani

et al. 2024

Preprint

View full text Add to dashboard Cite

Understanding complex interactions in biomedical networks is crucial for advancements in biomedicine. Traditional link prediction (LP) methods, using similarity metrics like Personalized PageRank, are limited in capturing the complexity of biological networks. Recently, representation-based learning techniques have emerged, mapping nodes to low-dimensional embeddings to enhance prediction accuracy. However, these methods often face challenges with interpretability and scalability in large, complex networks. Based on a representation of biological systems as knowledge graphs (KGs), which encode entities and their relationships as triplets, we propose here BioKGC, a novel graph neural network framework which builds upon the Neural Bellman-Ford Network (NBFNet). It addresses the limitations of previous methods by utilizing path-based reasoning for LP in biomedical knowledge graphs (KGs). Unlike node-embedding learning frameworks that optimize the embedding space based on single triplets, BioKGC learns representations between nodes by considering all relations along paths. This approach enhances prediction accuracy and interpretability, allowing for the visualization of influential paths and facilitating the validation of biological plausibility. BioKGC leverages a background regulatory graph (BRG) for enhanced message passing and implements a stringent negative sampling strategy to improve learning precision. In evaluations across various LP tasks — gene function annotation, drug-disease interaction prediction, synthetic lethality prediction, and lncRNA-mRNA regulatory relationship inference — BioKGC consistently outperformed state-of-the art methods. BioKGC outperformed knowledge graph embedding and GNN-based methods in gene function prediction, especially with BRG information. We demonstrated that BioKGC effectively predicts drug-disease interactions in zero-shot learning scenarios, surpassing state-of-the-art models like TxGNN. Additionally, BioKGC demonstrated robust performance in synthetic lethality prediction and the capacity for scoring novel lncRNA-mRNA interactions, showcasing its versatility in diverse biomedical applications. One of BioKGC’s key advantages is its interpretability, enabling researchers to trace prediction paths and gain insights into molecular mechanisms. Combined with its use of regulatory information for message passing, BioKGC is a powerful tool for predicting complex biological interactions, making it valuable for drug discovery and personalized medicine.

show abstract

Section: Lncrna-target Prediction Taskmentioning

confidence: 99%

Section: Lncrna-gene Target Prediction Taskmentioning

confidence: 99%