[Paper Review] Integrate Any Omics: Towards genome-wide data integration for patient stratification

Shihao Ma, Andy G.X. Zeng+4 more|arXiv (Cornell University)|Jan 15, 2024

Cancer Genomics and Diagnostics5 citations

TL;DR

IntegrAO is an unsupervised framework for integrating incomplete multi-omics profiles using partially overlapping patient graphs and graph neural networks, enabling subdivision of patients and accurate classification of new samples with partial data.

ABSTRACT

High-throughput omics profiling advancements have greatly enhanced cancer patient stratification. However, incomplete data in multi-omics integration presents a significant challenge, as traditional methods like sample exclusion or imputation often compromise biological diversity and dependencies. Furthermore, the critical task of accurately classifying new patients with partial omics data into existing subtypes is commonly overlooked. To address these issues, we introduce IntegrAO (Integrate Any Omics), an unsupervised framework for integrating incomplete multi-omics data and classifying new samples. IntegrAO first combines partially overlapping patient graphs from diverse omics sources and utilizes graph neural networks to produce unified patient embeddings. Our systematic evaluation across five cancer cohorts involving six omics modalities demonstrates IntegrAO's robustness to missing data and its accuracy in classifying new samples with partial profiles. An acute myeloid leukemia case study further validates its capability to uncover biological and clinical heterogeneity in incomplete datasets. IntegrAO's ability to handle heterogeneous and incomplete data makes it an essential tool for precision oncology, offering a holistic approach to patient characterization.

Motivation & Objective

• Address incomplete multi-omics data without discarding samples or imputing values.
• Develop an unsupervised framework to fuse partially overlapping omics graphs into unified patient embeddings.
• Enable accurate classification of new patients with partial omics data into predefined subtypes.
• Demonstrate robustness across multiple cancer cohorts and omics modalities.
• Validate biological and clinical relevance through case studies and survival/drug-response analyses.

Proposed method

• Construct per-omics patient graphs using nodes as patients and edges as pairwise similarities.
• Iteratively fuse partially overlapping graphs to create fused omics-specific graphs.
• Use omics-specific graph neural network encoders to learn low-dimensional patient embeddings.
• Align embeddings across omics in a shared space and average to obtain integrated embeddings.
• Optionally fine-tune with an MLP head to enable supervised subtype prediction for new patients.
• Evaluate with simulation (partial overlaps) and real cancer cohorts, including AML, across six omics modalities.

Experimental results

Research questions

• RQ1 Can IntegrAO robustly integrate partially overlapping multi-omics data without discarding samples?
• RQ2 How well can IntegrAO classify new patients with incomplete omics profiles into predefined subtypes?
• RQ3 Do integrated subtypes reveal clinically and biologically meaningful heterogeneity beyond single-omics analyses?
• RQ4 Is IntegrAO scalable and effective across multiple cancer types and omics modalities?
• RQ5 What is the impact of missing data patterns on clustering quality and survival/drug-response associations?

Key findings

• IntegrAO outperforms NEMO and MSNE across simulated partial-overlap scenarios.
• IntegrAO identifies 12 biologically and clinically distinct AML subtypes by integrating mRNA, DNA methylation, and cell hierarchy data.
• Integrated subtypes show improved survival differentiation and greater clinical enrichment compared to single data types.
• Across five TCGA cancer types with partial data, IntegrAO more reliably achieves differential survival and clinical enrichment than competitors.
• In new-patient classification, IntegrAO outperforms MLP, SVM, XGBoost, Random Forest, and KNN across various omic combinations.
• IntegrAO enables modality-agnostic inference, maintaining strong performance with incomplete data.