Contextualizing gene expression with feature rich graph neural networks
Ho, S. S.; Mills, R.
Show abstract
Variation in gene expression arises from the interplay between chromatin architecture, epigenetic marks, transcription factor binding, and regulatory elements - none alone capture the full complexity of the regulatory landscape. We introduce Omics Graph Learning (OGL), a deep learning paradigm that integrates multi-omics data into a network of biological entities. Leveraging over 950 curated datasets, OGL achieves highly accurate gene expression prediction across 20 cell lines and tissues. Through > 81 million in-silico perturbations, we uncover tissue-specific patterns of feature utilization, reveal non-additive interactions among epigenetic marks, and identify regulatory elements with disproportionate influence on gene expression. Our findings highlight complex synergistic and antagonistic relationships among molecular features that vary by cellular context. By bridging predictive accuracy with biological interpretation, OGL provides a framework for deciphering multifaceted mechanisms that shape the transcriptome.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Dynamic network-guided CRISPRi screen reveals CTCF loop-constrained nonlinear enhancer-gene regulatory activity in cell state transitions 98%
- Benchmarking of deep neural networks for predicting personal gene expression from DNA sequence highlights shortcomings 97%
- Linking regulatory variants to target genes by integrating single-cell multiome methods and genomic distance 97%
Similar papers in this journal
- Interpretable deep learning reveals the sequence rules of Hippo signaling 97%
- Gene regulatory network inference from CRISPR perturbations in primary CD4+ T cells elucidates the genomic basis of immune disease 97%
- Robust enhancer-gene regulation identified by single-cell transcriptomes and epigenomes 97%
Similar papers in this journal
- CREaTor: zero-shot cis-regulatory pattern modeling with attention mechanisms 98%
- An interpretable bimodal neural network characterizes the sequence and preexisting chromatin predictors of induced TF binding 98%
- Evaluating the representational power of pre-trained DNA language models for regulatory genomics 98%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.