Pathway-enhanced Transformer-based model for robust enumeration of cell types from the cell-free transcriptome (preprint)

ABSTRACT

Analysis cell types origin of cell-free RNA can enhance the resolution of liquid biopsies, thereby deepening the understanding of molecular and cellular changes in development and disease processes. Existing deconvolution methods typically rely on meticulously curated gene expression profiles or employ deep neural network with vast and complex solution spaces that are difficult to interpret. These approaches fail to leverage the synergistic and co-expression effects among genes in biological signaling pathways, compromising their generalizability and robustness. To address this issue, we have developed 'Deconformer', a Transformer-based deconvolution model that integrates biological signaling pathways at the embedding stage. Compared to popular methods on multiple datasets, Deconformer demonstrates superior performance and robustness, and is capable of tracking the developmental process of the placenta. Additionally, Deconformer's self-attention mechanism has identified a connection between platelet activation, dependencies with other pathways, and the severity of COVID-19. We believe that Deconformer will enable and accelerate the precise analysis of a wide range of cell-free RNA, describing disease progression and severity from the perspectives of originating cell fractions and pathway dependencies.