PRAGA: Prototype-aware Graph Adaptive Aggregation for Spatial Multi-modal Omics Analysis

Xinlei Huang,Zhiqi Ma,Dian Meng,Yanran Liu,Shiwei Ruan,Qingqiang Sun,Xubin Zheng,Ziyue Qiao

2024-09-20

Abstract:Spatial multi-modal omics technology, highlighted by Nature Methods as an advanced biological technique in 2023, plays a critical role in resolving biological regulatory processes with spatial context. Recently, graph neural networks based on K-nearest neighbor (KNN) graphs have gained prominence in spatial multi-modal omics methods due to their ability to model semantic relations between sequencing spots. However, the fixed KNN graph fails to capture the latent semantic relations hidden by the inevitable data perturbations during the biological sequencing process, resulting in the loss of semantic information. In addition, the common lack of spot annotation and class number priors in practice further hinders the optimization of spatial multi-modal omics models. Here, we propose a novel spatial multi-modal omics resolved framework, termed PRototype-Aware Graph Adaptative Aggregation for Spatial Multi-modal Omics Analysis (PRAGA). PRAGA constructs a dynamic graph to capture latent semantic relations and comprehensively integrate spatial information and feature semantics. The learnable graph structure can also denoise perturbations by learning cross-modal knowledge. Moreover, a dynamic prototype contrastive learning is proposed based on the dynamic adaptability of Bayesian Gaussian Mixture Models to optimize the multi-modal omics representations for unknown biological priors. Quantitative and qualitative experiments on simulated and real datasets with 7 competing methods demonstrate the superior performance of PRAGA.

Genomics,Machine Learning

What problem does this paper attempt to address?

The paper aims to address key challenges in the integration of spatial multimodal omics data, particularly how to encode omics features from different modalities along with corresponding spatial information into a unified latent space. Specifically:

1. **Addressing the limitations of fixed K-nearest neighbor (KNN) graphs**: Existing methods primarily simulate feature correlations between sequencing points by constructing KNN graphs and generate comprehensive representations through graph neural networks. However, these methods overlook the disturbances introduced during biological sequencing that interfere with semantic relationships, leading to fixed KNN graphs failing to capture potential semantic relationships.

2. **Proposing a dynamic graph structure**: To overcome this issue, this paper proposes a dynamic modality-specific graph structure that reveals potential semantic relationships through cross-modal knowledge learning and reduces the impact of disturbances.

3. **Prototype contrastive learning**: Since the annotation and number of types of sequencing points are usually unknown in real scenarios, optimizing the modality-specific graph becomes challenging. Therefore, this paper proposes a dynamic prototype contrastive learning method that uses a Bayesian Gaussian mixture model to adaptively perceive the number of sequencing point types and optimize the learnable graph structure.

4. **Experimental validation**: Through quantitative and qualitative experiments on simulated and real datasets, the proposed PRAGA method demonstrates superior performance in handling spatial multimodal omics data. Experimental results show that PRAGA significantly outperforms existing methods on multiple evaluation metrics, particularly achieving notable improvements in F1-Score and normalized mutual information (NMI).