We examined the public GitHub repository and the two associated Zenodo records. They appear to provide the datasets, training/evaluation code, and metric files, but we could not locate the trained benchmark checkpoints or the complete frozen inference metadata.
Would it be possible to share, or point us to a download location for, the following materials?
The five fold-specific trained checkpoints for DeepLoc2-ProtT5, DeepLoc2-ESM1, MULocDeep, and LAProtT5, particularly the level-1 models;
If available, the five checkpoints for the selected ProtT5-MHA reference model;
The class order and fold-specific MCC thresholds (all_thresholds.npy), or the validation-fold prediction files needed to reproduce those thresholds exactly;
The corresponding configuration files, run IDs, preprocessing settings, and an inference script or manifest for predictions on arbitrary FASTA sequences;
Instructions for generating the required PLM embeddings consistently, if the precomputed embedding files cannot be shared;
Any applicable license, citation, or usage requirements.
A Zenodo deposit, institutional storage link, or temporary download link would all work for us. We would, of course, cite the article and associated resources appropriately.
Thank you very much for your help and for making the benchmark datasets and code publicly available.
We examined the public GitHub repository and the two associated Zenodo records. They appear to provide the datasets, training/evaluation code, and metric files, but we could not locate the trained benchmark checkpoints or the complete frozen inference metadata.
Would it be possible to share, or point us to a download location for, the following materials?
The five fold-specific trained checkpoints for DeepLoc2-ProtT5, DeepLoc2-ESM1, MULocDeep, and LAProtT5, particularly the level-1 models;
If available, the five checkpoints for the selected ProtT5-MHA reference model;
The class order and fold-specific MCC thresholds (all_thresholds.npy), or the validation-fold prediction files needed to reproduce those thresholds exactly;
The corresponding configuration files, run IDs, preprocessing settings, and an inference script or manifest for predictions on arbitrary FASTA sequences;
Instructions for generating the required PLM embeddings consistently, if the precomputed embedding files cannot be shared;
Any applicable license, citation, or usage requirements.
A Zenodo deposit, institutional storage link, or temporary download link would all work for us. We would, of course, cite the article and associated resources appropriately.
Thank you very much for your help and for making the benchmark datasets and code publicly available.