BOOM (Benchmarking Out-Of-distribution Molecular Property Predictions)
收藏资源简介:
BOOM数据集是一个用于评估机器学习模型在分子性质预测任务中泛化到分布外(OOD)性能的标准基准。该数据集由10个独特的分子性质数据集组成,包括QM9数据集中的8个分子性质和10k数据集中的2个分子性质。QM9数据集包含133,886个小分子,而10k数据集包含10,206个实验合成的CHON小分子。这些数据集通过密度泛函理论(DFT)计算获得。BOOM数据集旨在解决当前化学机器学习模型在分布外泛化能力不足的问题,并推动开发具有更强泛化能力的化学基础模型。
The BOOM dataset is a standard benchmark for evaluating the out-of-distribution (OOD) generalization performance of machine learning models on molecular property prediction tasks. It consists of 10 unique molecular property datasets, including 8 molecular properties from the QM9 dataset and 2 from the 10k dataset. The QM9 dataset contains 133,886 small molecules, while the 10k dataset comprises 10,206 experimentally synthesized CHON small molecules. All these datasets are obtained via density functional theory (DFT) calculations. The BOOM dataset aims to address the insufficient out-of-distribution generalization capability of current chemical machine learning models, and promote the development of more robust chemical foundation models.




