SIGMA: Structure-Invariant Generative Molecular Alignment for Chemical Language Models via Autoregressive Contrastive Learning

arXiv cs.LG / 3/27/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses a key problem in string-based autoregressive molecular generation: the same molecular graph can correspond to multiple token sequences, causing latent “trajectory divergence” as linearization history changes representations of equivalent partial graphs.
It introduces Structure-Invariant Generative Molecular Alignment (SIGMA), which keeps linear string representations but uses a token-level contrastive objective to align latent states for prefixes that are consistent with identical suffixes while respecting geometric/structural symmetries.
To improve inference efficiency and avoid redundant exploration, the authors propose Isomorphic Beam Search (IsoBeam), which prunes isomorphic-equivalent paths dynamically during decoding.
Experiments on standard benchmarks indicate SIGMA improves the balance between sequence scalability and graph fidelity, achieving better sample efficiency and structural diversity during multi-parameter optimization versus strong baselines.

Abstract

Linearized string representations serve as the foundation of scalable autoregressive molecular generation; however, they introduce a fundamental modality mismatch where a single molecular graph maps to multiple distinct sequences. This ambiguity leads to \textit{trajectory divergence}, where the latent representations of structurally equivalent partial graphs drift apart due to differences in linearization history. To resolve this without abandoning the efficient string formulation, we propose Structure-Invariant Generative Molecular Alignment (SIGMA). Rather than altering the linear representation, SIGMA enables the model to strictly recognize geometric symmetries via a token-level contrastive objective, which explicitly aligns the latent states of prefixes that share identical suffixes. Furthermore, we introduce Isomorphic Beam Search (IsoBeam) to eliminate isomorphic redundancy during inference by dynamically pruning equivalent paths. Empirical evaluations on standard benchmarks demonstrate that SIGMA bridges the gap between sequence scalability and graph fidelity, yielding superior sample efficiency and structural diversity in multi-parameter optimization compared to strong baselines.