Representation Learning with Blockwise Missingness and Signal Heterogeneity

TL;DR

This paper introduces APPCA, a robust representation learning method for multi-source data with blockwise missingness and signal heterogeneity, improving embedding accuracy in complex real-world scenarios.

Contribution

We develop APPCA, a novel framework that effectively handles structured blockwise missingness and heterogeneity, with theoretical guarantees and practical validation.

Findings

APPCA outperforms existing methods in simulations.

APPCA achieves accurate embeddings in single-cell sequencing data.

Theoretical bounds show robustness to signal heterogeneity.

Abstract

Unified representation learning for multi-source data integration faces two important challenges: blockwise missingness and blockwise signal heterogeneity. The former arises from sources observing different, yet potentially overlapping, feature sets, while the latter involves varying signal strengths across subject groups and feature sets. While existing methods perform well with fully observed data or uniform signal strength, their performance degenerates when these two challenges coincide, which is common in practice. To address this, we propose Anchor Projected Principal Component Analysis (APPCA), a general framework for representation learning with structured blockwise missingness that is robust to signal heterogeneity. APPCA first recovers robust group-specific column spaces using all observed feature sets, and then aligns them by projecting shared "anchor" features onto these subspaces before performing PCA. This projection step induces a significant denoising effect. We establish estimation error bounds for embedding reconstruction through a fine-grained perturbation analysis. In particular, using a novel spectral slicing technique, our bound eliminates the standard dependency on the signal strength of subject embeddings, relying instead solely on the signal strength of integrated feature sets. We validate the proposed method through extensive simulation studies and an application to multimodal single-cell sequencing data.