APARENT2 Training Data and Models

Processed training data for the APARENT2 model (measurements from the random MPRA and designed oligo pool originally published by Bogard et al., 2019; see https://doi.org/10.1016/j.cell.2019.04.046 for reference). This repository also contains the APARENT2 model file. For more information on the training procedure, see the Genome Biology article "Deciphering the impact of genetic variation on human polyadenylation using APARENT2" (https://genomebiology.biomedcentral.com/articles/10.1186/s13059-022-02799-4). Two versions of the model are available:

(a) aparent_all_libs_resnet_no_clinvar_wt_ep_5.h5: The originally trained APARENT2 model.
(b) aparent_all_libs_resnet_no_clinvar_wt_ep_5_var_batch_size_inference_mode_no_drop.h5: Identical weights and predictions as model (a), but the normalization layers have been set to inference mode and the dropout layers have been removed (thus making it compatible with the scrambler pipeline).

本数据集为APARENT2模型的预处理训练数据，数据来源于Bogard等人2019年首次发表的随机大规模平行报告分析（random MPRA）与设计寡核苷酸池（designed oligo pool），相关参考文献可参见https://doi.org/10.1016/j.cell.2019.04.046。本仓库同时包含APARENT2模型文件。有关模型训练流程的详细信息，请参阅《Genome Biology》期刊论文《利用APARENT2解析遗传变异对人类多聚腺苷酸化的影响》，链接为https://genomebiology.biomedcentral.com/articles/10.1186/s13059-022-02799-4。当前提供两款模型版本：
(a) aparent_all_libs_resnet_no_clinvar_wt_ep_5.h5：初始训练得到的原生APARENT2模型。
(b) aparent_all_libs_resnet_no_clinvar_wt_ep_5_var_batch_size_inference_mode_no_drop.h5：该模型与(a)款模型的权重及预测结果完全一致，但将归一化层设置为推理模式并移除了随机失活（Dropout）层，因此可与scrambler流水线兼容。