sEH dataset for GFlowNet.

Total unique molecule graphs = 34,012,224. Reward is from a proxy model trained to predict binding affinity to soluble epoxide hydrolase from AutoDock. We build using 18 blocks with 2 stems, and use 6 blocks per molecule. <br> block_18_stop6.pkl.gz is a dictionary mapping x (strings) to AutoDock rewards. sehstr_gbtr_allpreds.pkl.gz is a dictionary mapping x (strings) to proxy model rewards.

总唯一分子图（molecule graphs）总计达34,012,224张。奖励值源自一款经训练的代理模型（proxy model），该模型可通过AutoDock工具预测分子与可溶性环氧化物水解酶（soluble epoxide hydrolase）的结合亲和力。本次数据集构建采用18个带2个连接臂的结构块，且每个分子需使用6个结构块。
block_18_stop6.pkl.gz为一个字典，其键为字符串形式的x，值为AutoDock奖励值；sehstr_gbtr_allpreds.pkl.gz同样为一个字典，其键为字符串形式的x，值为代理模型生成的奖励值。