Paper List

Journal: ArXiv Preprint
Published: Unknown
BioinformaticsSystems Biology

Hierarchical Molecular Language Models (HMLMs)

Department of Chemical Engineering, University of Arkansas, Fayetteville, AR 72701, USA

Hasi Hays, Yue Yu, William Richardson
Figure
Figure
Figure
Figure

The 30-Second View

IN SHORT: This paper addresses the core challenge of accurately modeling context-dependent signaling, pathway cross-talk, and temporal dynamics across multiple biological scales in cellular signaling networks.

Innovation (TL;DR)

  • Methodology Introduces cellular signaling as a molecular language with unique grammar and semantics, establishing a theoretical foundation for molecular artificial intelligence (MAI).
  • Methodology Develops HMLMs as a novel computational architecture adapting transformer architecture to model signaling networks as information-processing systems across molecular, pathway, and cellular scales.
  • Methodology Implements graph-structured attention mechanisms and hierarchical scale-bridging operators (aggregation, decomposition, translation) to accommodate signaling network topology and multi-scale organization.

Key conclusions

  • HMLM achieved MSE of 0.058 for temporal signaling predictions, representing 30% improvement over GNNs (0.083) and 52% improvement over ODE models (0.121).
  • Under sparse temporal sampling with only 4 timepoints, HMLM maintained superior performance with MSE = 0.041, demonstrating robustness to limited temporal data.
  • Attention mechanisms identified biologically plausible pathway interactions including mechanotransduction-MAPK coupling and TGFβ to ERK signaling, validating the model's ability to capture meaningful biological relationships.
Background and Gap： Traditional modeling approaches (ODEs, Boolean networks, Bayesian networks) fail to adequately capture context-dependent signaling, pathway cross-talk, and temporal dynamics across multiple biological scales, particularly under sparse temporal sampling conditions.

Abstract: Cellular signaling networks represent complex information processing systems that have been modeled via traditional mathematical or statistical approaches. However, these methods often struggle to capture context-dependent signaling, pathway cross-talk, and temporal dynamics across multiple biological scales. Here, we introduce hierarchical molecular language models (HMLMs), a novel architecture that proposes a molecular network-specifiac large language model (LLM) to use in intracellular communication as a specialized molecular language, which includes molecules as tokens, protein interactions, post-translational modifications, and regulatory events modeled as semantic relationships within an adapted transformer architecture. HMLMs employ graph-structured attention mechanisms to accommodate signaling network topology while integrating information across the molecular, pathway, and cellular scales through hierarchical attention patterns. We demonstrate HMLM superiority using a cardiac fibroblast signaling network comprising over 100 molecular species across functional modules connected by regulatory edges. HMLM achieved a mean squared error (MSE) of 0.058 for temporal signaling predictions, representing 30% improvement over graph neural networks (GNNs: 0.083) and 52% improvement over ordinary differential equation models (ODEs: 0.121), with particular advantages under sparse temporal sampling conditions where HMLM maintained MSE = 0.041 with only 4 timepoints. The attention-based computational analysis identified key inter-pathway cross-talk patterns through learned attention mechanisms, including mechanotransduction-MAPK coupling and TGFβ to ERK signaling, demonstrating the model's capability to capture biologically plausible pathway interactions from network topology and temporal dynamics and convergent regulatory mechanisms controlling fibrosis markers in simulated cardiac fibroblast networks. The HMLMs offer a foundation for AI-driven biology and medicine with predictable scaling characteristics suitable for interactive applications. By bridging molecular mechanisms with cellular phenotypes through AI-driven molecular language representation, HMLMs provide a powerful paradigm for systems biology that advances precision medicine applications and therapeutic discovery in the era of AI.