MOES-Pred: Molecular Structural Representation Learning by Adaptive Energy-Sentinel Vibration for Generalized Property Prediction

Predicting molecular properties from three-dimensional structures is fundamentally hindered by limited labeled data. While researchers have adapted self-supervised pre-training techniques from computer vision and natural language processing to address this scarcity, these approaches frequently neglect the intrinsic physical principles unique to molecular systems. From a physical perspective, denoising pre-training can be formally proven equivalent to learning molecular force fields. However, existing methods indiscriminately apply uniform noise across all molecules, thereby introducing systematic bias into the modeling of molecular distributions. To mitigate this issue, we introduce MOES-Pred, a denoising pre-training framework featuring an energy sentinel mechanism that dynamically tailors noise perturbations to individual molecules. Leveraging chemical prior knowledge, our molecule-specific noising strategies enhance conformational sampling coverage and improve distribution modeling fidelity. Extensive experiments show that MOES-Pred surpasses mainstream approaches in both force prediction and downstream quantum chemical property prediction, demonstrating substantial improvements.

Computational models struggle to accurately predict molecular behaviors due to limited reliable molecular data. Existing artificial intelligence tools borrowed from other research fields ignore the natural physical properties of molecules, resulting in unreliable prediction results. Adding and removing random noise is a common way to improve model learning. However, current methods use the same noise for all molecules, which causes persistent errors and limits model effectiveness.We introduce MOES-Pred, a new learning tool that adjusts noise uniquely for each molecule following basic chemical rules. This customized approach helps the model understand molecules better. Our tests confirm that MOES-Pred outperforms existing tools, delivering more accurate molecular predictions for scientific research.