----------------------------------------------------------Record 4/4 — output.tar.part-12 to output.tar.part-16 (~43.3 GB)----------------------------------------------------------Part 4 of 4 of the AlphaFold 3 model dataset accompanying Toghani, Seager et al., "AI-guided discovery of atypical protein assemblies." The full dataset (~164 GB) is a single tar archive (output.tar) split into 17 ten-gigabyte parts, distributed across four sibling Zenodo records to satisfy the 50 GB per-record limit. This record contains parts 12–16 of 16 (five parts, including the final 3.3 GB tail). To use the data you must also download the parts from the other three sibling records together with the README, SHA-256 checksums and reassembly script from the index record [DOI: https://doi.org/10.5281/zenodo.19860918]; place all 17 parts in the same folder and run reassemble.sh. License: CC0 1.0.