Learning Cell-Aware Hierarchical Multi-Modal Representations for Robust Molecular Modeling

2025-11-26

AI Summary

Existing cell perception methods face two key challenges: incomplete external biological data modalities and difficulty in modeling hierarchical dependencies across molecular, cellular, and genomic scales. To address these, we propose CHMR—a novel framework that explicitly encodes multi-scale hierarchical relationships via a tree-structured vector quantization (VQ) module. CHMR integrates chemical structures with cellular phenotypes (morphology and gene expression) to construct robust multimodal representations. By unifying multimodal deep learning with hierarchical vector quantization, it jointly optimizes local and global cross-modal dependencies. Evaluated on nine benchmarks comprising 728 tasks, CHMR achieves average improvements of 3.6% in classification accuracy and 17.2% in regression performance over state-of-the-art methods. Its core contribution lies in the first incorporation of biologically grounded hierarchical priors into a vector quantization architecture, enabling end-to-end, cross-scale joint representation learning.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Understanding how chemical perturbations propagate through biological systems is essential for robust molecular property prediction. While most existing methods focus on chemical structures alone, recent advances highlight the crucial role of cellular responses such as morphology and gene expression in shaping drug effects. However, current cell-aware approaches face two key limitations: (1) modality incompleteness in external biological data, and (2) insufficient modeling of hierarchical dependencies across molecular, cellular, and genomic levels. We propose CHMR (Cell-aware Hierarchical Multi-modal Representations), a robust framework that jointly models local-global dependencies between molecules and cellular responses and captures latent biological hierarchies via a novel tree-structured vector quantization module. Evaluated on nine public benchmarks spanning 728 tasks, CHMR outperforms state-of-the-art baselines, yielding average improvements of 3.6% on classification and 17.2% on regression tasks. These results demonstrate the advantage of hierarchy-aware, multimodal learning for reliable and biologically grounded molecular representations, offering a generalizable framework for integrative biomedical modeling. The code is in https://github.com/limengran98/CHMR.

Problem

• Modeling hierarchical dependencies across molecular, cellular, and genomic levels
• Addressing modality incompleteness in external biological data sources
• Integrating cellular responses with chemical structures for drug effects

Innovation

• Jointly models local-global dependencies between molecules and cells
• Captures latent biological hierarchies via tree-structured vector quantization
• Integrates multimodal cellular responses for robust molecular representations