HiAntitoxinPred artifacts v2

This record provides reproducibility artifacts for HiAntitoxinPred, a protein language model-based framework for bacterial antitoxin prediction. The archive includes the released pretrained checkpoint, the full five-seed checkpoint set, and the ESM2-1280 per-residue feature files for the main benchmark, required to reproduce the main evaluation results. The processed benchmark datasets and documentation are maintained in the accompanying GitHub repository. The main benchmark uses a 1:10 class-imbalanced, length-stratified split with protein sequences capped at 200 amino acids. The released checkpoint corresponds to the locked main benchmark seed-42 model and reproduces the reported test-set performance when evaluated with the validation-selected threshold protocol described in the accompanying GitHub repository. These files are intended to be used together with the HiAntitoxinPred source code and documentation: https://github.com/chelseawhu/HiAntitoxinPred