SciCore-Mol: Augmenting Large Language Models with Pluggable Molecular Cognition Modules

2026-05-21

AI Summary

This work addresses the challenge that large language models (LLMs) struggle to reason effectively over heterogeneous scientific data such as molecules, primarily due to the semantic gap between discrete textual symbols and continuous or topological chemical representations. To bridge this gap, the authors propose a modular, plug-and-play cognitive architecture comprising three core components: topology-aware encoding, latent diffusion-based generation, and reaction-aware reasoning. These modules are deeply integrated with an LLM through learnable interfaces, systematically endowing it with molecular-level expertise. Moving beyond conventional text-centric paradigms, the framework achieves substantial performance gains across molecular understanding, generation, reaction prediction, and knowledge synthesis tasks. The resulting 8B-parameter open-source system matches or even surpasses leading closed-source LLMs on multiple benchmarks.

Abstract

Large Language Models (LLMs) are central to the one-for-all intelligent paradigm, but they face a fundamental challenge when dealing with heterogeneous scientific data such as molecules: the inherent gap between discrete linguistic symbols and topological molecular or continuous reaction data leads to significant information loss and semantic noise in text-based reasoning. We propose SciCore-Mol, a modular framework that bridges this gap through three deeply integrated pluggable cognitive modules: a topology-aware perception module, a latent diffusion-based molecular generation module, and a reaction-aware reasoning module. Each module is coupled to the LLM backbone through learned representation interfaces, enabling richer information exchange than is possible with text-only tool feedback. Our experiments on diverse chemical tasks demonstrate that SciCore-Mol achieves strong comprehensive performance across molecular understanding, generation, reaction prediction, and general chemistry knowledge, with an 8B-parameter open-source system that is competitive with and in several dimensions surpasses proprietary large models. This work provides a systematic blueprint for equipping LLMs with scientific expertise through decoupled, pluggable, and flexibly orchestrated modules, with direct implications for drug design, chemical synthesis, and broader scientific discovery.

Problem

• Large Language Models
• Molecular Data
• Heterogeneous Scientific Data
• Semantic Noise
• Information Loss

Innovation

• pluggable cognitive modules
• molecular representation
• latent diffusion
• reaction-aware reasoning
• LLM augmentation