SymSpectra: Symmetric Information Bottleneck Framework for Molecular Structure Recognition under Imbalanced Settings

Identifying molecular structures from spectral data is essential for early-stage chemical analysis, yet it remains a difficult task due to severe functional group imbalance and complex inter-group dependencies, which often cause existing methods to overfit frequent groups while underperforming on rare ones. To address these issues, we present SymSpectra, a Symmetric Conditional Information Bottleneck (SCIB) framework designed to seamlessly integrate multi-modal Spectra features. Our model employs the SCIB framework to fuse multi-modal spectroscopic data into a unified representation, effectively preserving discriminative signals while mitigating redundancy. To enhance robustness against data imbalance, we incorporate conditional mutual information into the training objective, increasing the model’s sensitivity to rare functional groups and challenging molecular cases. Additionally, a specialized module captures the dependencies among functional groups, improving both prediction accuracy and chemically meaningful interpretability. Experiments on multimodal spectral datasets show that SymSpectra outperforms state-of-the-art methods, achieving an F1-score of 0.970 in substructure classification and demonstrating strong robustness under various imbalance settings.

To discover new drugs or materials, scientists must figure out a molecule's exact structure using various spectral data. While AI can help analyze these complex signals, it often struggles because some molecular substructures are extremely common, while others are rare. As a result, standard AI models tend to ignore the rare yet often crucial pieces and fail to understand how different molecular parts naturally fit together. To solve this, we developed SymSpectra, an AI framework that seamlessly combines different types of multi-modal spectra into a single, clear picture. Instead of getting distracted by the most frequent molecular parts, our tool uses a specialized mathematical framework to actively prioritize and identify rare components. We also designed it to learn the fundamental rules of chemistry, allowing it to understand which molecular building blocks logically co-occur. By overcoming these data imbalances, SymSpectra identifies chemical structures with high accuracy. This will allow researchers to reliably discover and design novel molecules for medicine and advanced materials.