DOGMA: Weaving Structural Information into Data-centric Single-cell Transcriptomics Analysis

TL;DR

DOGMA reframes single-cell transcriptomics as a data-centric problem, arguing that high model capacity cannot compensate for noisy data and missing biological structure. It introduces a deterministic, ontology-guided framework that fuses multi-level priors—Cell Ontology, Gene Ontology, and Phylogeny—into a robust cell graph and semantic feature space, enabling plug-and-play use with standard GNNs. Across cross-species and multi-organ benchmarks, DOGMA achieves state-of-the-art performance, strong zero-shot generalization, and substantial data and compute efficiency, outperforming sequence-based and other graph-based baselines. This work demonstrates that explicit incorporation of biological structure can yield more reliable, interpretable, and scalable single-cell analyses than simply scaling model size.

Abstract

Recently, data-centric AI methodology has been a dominant paradigm in single-cell transcriptomics analysis, which treats data representation rather than model complexity as the fundamental bottleneck. In the review of current studies, earlier sequence methods treat cells as independent entities and adapt prevalent ML models to analyze their directly inherited sequence data. Despite their simplicity and intuition, these methods overlook the latent intercellular relationships driven by the functional mechanisms of biological systems and the inherent quality issues of the raw sequence data. Therefore, a series of structured methods has emerged. Although they employ various heuristic rules to capture intricate intercellular relationships and enhance the raw sequencing data, these methods often neglect biological prior knowledge. This omission incurs substantial overhead and yields suboptimal graph representations, thereby hindering the utility of ML models. To address them, we propose DOGMA, a holistic data-centric framework designed for the structural reshaping and semantic enhancement of raw data through multi-level biological prior knowledge. Transcending reliance on stochastic heuristics, DOGMA redefines graph construction by integrating Statistical Anchors with Cell Ontology and Phylogenetic Trees to enable deterministic structure discovery and robust cross-species alignment. Furthermore, Gene Ontology is utilized to bridge the feature-level semantic gap by incorporating functional priors. In complex multi-species and multi-organ benchmarks, DOGMA achieves SOTA performance, exhibiting superior zero-shot robustness and sample efficiency while operating with significantly lower computational cost.

Figures (4)

• Figure 1: DOGMA: A Data-Centric Paradigm Shift.(Left) Data structure of current input paradigms. Sequence-based inputs lose topological structure, while heterogeneous graphs suffer from structural redundancy. (Middle) DOGMA acts as a universal structure inference engine. It injects multi-level knowledge (Gene Ontology, Cell Ontology, Phylogeny) to produce a knowledge-guided cell graph. (Right) Empirical validation. (a) GNN achieves the Target Zone (High Efficiency & High Performance) with significantly fewer parameters than Transformers. (b) DOGMA overcomes topological bottlenecks, reducing memory usage by 10x compared to scMoGNN while maintaining competitive accuracy.
• Figure 2: The DOGMA Framework(Top) Traditional pipelines rely on black-box models that lack prior knowledge, leading to high computational costs and suboptimal performance. (Middle) Our proposed data-centric workflow transforms raw scRNA-seq data into a knowledge-guided cell graph through six stages: (a-b) rigorous data curation and quality control; (c-d) representative downsampling and dimensionality reduction; (e) deterministic topology construction integrating Phylogeny, Cell Ontology, and MNN edges; and (f) feature enhancement via Gene Ontology (GO). (Bottom) The resulting Cell Graph serves as a universal, interpretable input for Graph Models, enabling high-performance, transferable analysis across zero-shot and clustering tasks.
• Figure 3: Data Efficiency Analysis (Q3). Performance comparison under varying training ratios on the Multi dataset. DOGMA (Red) maintains high accuracy even in data-scarce regimes, demonstrating superior sample efficiency compared to scPriorGraph, scMoGNN and scGPT
• Figure 4: Computational Efficiency Analysis (Q4). Comparison of inference time, RAM usage, and GPU memory consumption. DOGMA operates with significantly lower resource overhead, achieving both accuracy and efficiency.