Unveiling causal regulatory mechanisms through cell-state parallax

Area of Science:

• Genomics
• Computational Biology
• Neuroscience

Background:

• Genome-wide association studies (GWAS) identify disease-associated genetic variants in noncoding regions.
• Understanding the tissue-specific regulatory roles of these variants is crucial for therapeutic development.
• Current computational methods struggle with variant-level precision and causal inference in single-cell data.

Purpose of the Study:

• To introduce GrID-Net, a graph neural network approach for inferring causal locus-gene associations from single-cell multimodal data.
• To leverage the concept of "cell-state parallax" to infer causal mechanisms from static single-cell snapshots.
• To identify noncoding regulatory mechanisms underlying schizophrenia (SCZ).

Main Methods:

• Developed GrID-Net, a graph neural network generalizing Granger causal inference for single-cell trajectories.
• Applied GrID-Net to single-cell chromatin accessibility and gene expression data.
• Utilized a
• cell-state parallax
• approach to infer causal relationships from time-lagged epigenetic and transcriptional states.

Main Results:

• GrID-Net increased variant coverage by 36% for SCZ genetic variants.
• Identified noncoding mechanisms dysregulating 132 genes, including potassium transporters KCNG2 and SLC12A6.
• Discovered the significant role of neural transcription-factor binding disruptions in SCZ etiology.

Conclusions:

• GrID-Net provides a strategy for elucidating the tissue-specific impact of noncoding variants.
• The "cell-state parallax" concept offers a breakthrough for discovering gene regulatory mechanisms in single-cell multiomics.
• This approach advances the understanding of genetic contributions to complex diseases.