# DAGFormer: A graph-based domain adaptation approach for single-cell cancer drug response prediction

## TL;DR

DAGFormer is a new method that combines bulk and single-cell RNA data to better predict cancer drug responses at the single-cell level.

## Contribution

DAGFormer introduces a graph-based domain adaptation framework that captures intercellular interactions and reduces batch effects between bulk and single-cell RNA data.

## Key findings

- DAGFormer outperforms existing methods in predicting single-cell drug responses.
- The framework effectively reduces distribution gaps between bulk and single-cell RNA data.
- Cellular neighbor graphs improve the accuracy of drug response predictions.

## Abstract

Developing computational methods for single-cell drug response prediction deepens our understanding of tumor heterogeneity and uncovers resistance mechanisms critical to improving cancer therapy. However, current approaches struggle to fully capture intratumoral heterogeneity, as bulk RNA sequencing (bulk RNA-seq) obscures heterogeneity across individual cells, while single-cell RNA sequencing (scRNA-seq) remains constrained by limited throughput and high cost. Current approaches integrating bulk and scRNA-seq data frequently encounter batch effects, impairing robust knowledge transfer. Moreover, most existing methods overlook the role of intercellular interactions, treating cells as isolated entities. To overcome these limitations, we propose DAGFormer, a Graph-based Domain Adaptation framework that integrates bulk and scRNA-seq data for predicting single-cell drug responses. DAGFormer constructs cellular neighbor graphs using diverse topological strategies and employs Graph Domain Adaptation (GDA) to bridge graph-level distribution gaps between bulk and single-cell RNA-seq data. A dual-domain decoder further disentangles shared and modality-specific representations, preserving both general and unique biological signals. Benchmarking DAGFormer on ten independent scRNA-seq datasets demonstrated its superior performance compared to existing methods, underscoring its effectiveness and robustness in cancer drug response prediction.

In the era of precision medicine, single-cell RNA sequencing (scRNA-seq) provides gene expression profiles at the individual cell level, while bulk RNA sequencing (bulk RNA-seq) captures the averaged transcriptome of mixed cell populations. Predicting drug sensitivities at the single-cell level can offer valuable insights into the mechanisms of treatment response heterogeneity and drug resistance. Few studies to date have successfully integrated bulk RNA-seq and scRNA-seq data for drug response prediction, and only a limited number of computational methods have achieved promising results. However, this computational approach has a key limitation: drug response at the single-cell level is not solely determined by the intrinsic gene expression of individual cells, but is also influenced by intercellular interactions within the tumor microenvironment. To overcome the above limitation, we propose DAGFormer, a novel computational framework. The core idea of DAGFormer is to construct cellular neighbor graphs using different topological strategies and apply Graph Domain Adaptation (GDA) to reduce the distributional gap between cell relationship graphs derived from bulk and single-cell RNA-seq data. The experimental results show that by constructing cellular relationship graphs, DAGFormer effectively addresses the inherent batch effects between cell graphs derived from bulk and single-cell RNA-seq data, and accurately predicts cancer drug responses at the single-cell level.