inDecay model weights: cross-species transfer-learning checkpoints (Figures 4 & 5)

PyTorch Lightning fine-tuned checkpoints for reproducing Figures 4 and 5 of the inDecay manuscript (CRISPR indel-outcome prediction with cross-species transfer learning). Contents Fig4_mouse_checkpoint.tar.gz — per-gene mouse mESC fine-tuned models (52-fold). Extract to pl_trainer_log/True_mESC_featv5_c20_ST_DeepDecay_mul_identity_lr0.0003_L20.4_T0.5/mouse/. Fig5_livestock_checkpoint.tar.gz — per-species cross-species fine-tuned models for mESC, iPSC, K562, HAP1, CHO (12-fold). Extract to pl_trainer_log/True_{cell}_featv5_c20_ST_DeepDecay_mul_identity_lr0.0003_L20.5_T0.5/species/. pretrained_backbones.tar.gz — pretrained cell-line backbone checkpoints (CHO, HAP1, K562, iPSC, mESC; featv5/featv2 variants) that generated the Figure 2 benchmarking and Figure 3 transfer-learning results. Extract to pretrained/. Consumed by notebooks/Figures4_and_5.ipynb in the inDecay code repository (https://github.com/StatBiomed/inDecay) via scripts/fetch_data.py. Companion training / somatic data are archived on figshare (DOI 10.6084/m9.figshare.25133564).