MTtrans Model weight

About this figshare deposit:This figshare item is to store the weights of trained models. For source code and reproducibility about the study, please visit our github repository: https://github.com/holab-hku/MTtrans.

The naming of classes of model: M stands for MPRA (Sample et al). R stands for ribosome profiling (Cao et al). As MTtrans is a multi-task model which aims to integrate different techniques, different combination of M and R leads to many different model classes. For example, 3M means using 3 MPRA datasets to collectively train one MTtrans model, the resulting model will have 3 GRU tower. sMR means a sub-sampled MPRA was used together with a RP dataset.Model checkpoints and configsfiles name after .pth is the pytorch checkpoint file and _cv%d stands for different training repeats. Those name after .ini are the configuration files recording the hyper parameters and all other training settings.
small_repective_filed_strides1113-model_best_cvx.pth : the 3M model in the paper. The training process of model involves 3 MPRA datasets and V_PC3_RMSprop-model_best_cvx.pth: model sMR in the paper, which integrates a sub-sampled MRPA and the PC3 datasetsV_muscle_RMSprop-model_best_cvx.pth : sMR model which use muscle RP dataset.V_293_RMSprop-model_best_cvx.pth : sMR model which trains on HEK293 RP dataset.