Hratch Baghdassarian

This paper introduces a method to evaluate the robustness of biology-informed neural networks by measuring their ability to prune spurious interactions during training.

scLEMBAS models signaling pathway activity at single-cell resolution using a context-aware neural network to predict perturbation responses and capture cell-type-specific interactions.

The paper develops human ME-Models that integrate gene expression and metabolism, showing improved growth-rate predictions and biological plausibility in cancer cell lines.

Tensor-cell2cell v2 jointly analyzes protein- and metabolite-mediated cell-cell communication over time, revealing coordinated dynamics in brain organoid development.

The paper presents secRecon, a comprehensive secretory pathway reconstruction, and uses it to identify secretory differences in plasma cells and CHO cells, highlighting potential targets for improving secretion capacity.

The paper shows that integrating mRNA and protein data improves understanding of metastasis by leveraging complementary and reinforcing signals between modalities.

The paper uses single-cell sequencing to analyze CAR signaling and identifies ICD architectures that promote durable in vivo persistence in liquid tumors.

This paper integrates LIANA and Tensor-cell2cell to analyze cell-cell communication across multiple samples, enabling robust and flexible identification of communication programs.

The paper introduces a computational framework to infer the spatial code of cell-cell interactions using transcriptomic data from C. elegans, validated with experimental confirmation.

Tensor-cell2cell improves cell-cell communication analysis by incorporating cellular context, identifying communication modules linked to diseases like COVID-19 and Autism Spectrum Disorder.

The paper connects Alzheimer's disease genetics to amyloid formation by analyzing secretory pathway dysregulation in AD brains.

This study quantifies how repeated freeze-thaw cycles affect RNA-Seq reliability, showing that poly(A)-enriched samples lose reproducibility after three cycles.