HAM(Human-annotated Mappings)

HAM 是用于分子图分区的数据集。该数据集包含 1206 个重原子少于 25 个的有机分子的粗粒度 (CG) 映射。每个分子从 PubChem 数据库下载为 SMILES。一个分子被分配给两个注释器，以比较 CG 映射之间的人类一致性。下载的微笑是手工绘制的。完成的注释由第三人审查，以识别和删除不符合给定指南的不合理映射（例如：一个珠映射）。因此，当前数据库中每个分子有 1.68 个注释（删除了 16%）。

HAM is a dataset for molecular graph partitioning. It contains coarse-grained (CG) mappings for 1206 organic molecules with fewer than 25 heavy atoms. Each molecule was retrieved from the PubChem database in the form of a SMILES string. Two annotators were assigned to each molecule to evaluate the inter-annotator consistency of their respective CG mappings. The downloaded SMILES strings were manually prepared. The completed annotations were reviewed by a third party to identify and remove invalid mappings that fail to meet the given guidelines, such as one-bead mappings. As a result, each molecule in the current dataset has an average of 1.68 annotations, with 16% of the initial annotations eliminated.

搜集汇总

背景与挑战

背景概述

HAM是一个用于分子图分区的数据集，包含1206个重原子少于25个的有机分子的粗粒度映射。每个分子由两名注释器标注并经过审查，平均每个分子有1.68个有效注释。