Data and code for plant-cell-wall-degrading enzymes in ectomycorrhizal fungi: gene-family reduction, labelling design and type I error, and RELAX reproducibility

Analysis code, derived data and result files underlying a study of plant-cell-wall-degrading enzymes in ectomycorrhizal fungi. Contents. Two archives. aicomp-h1h3-deposit_v1.5.zip is the main deposit: 1,778 files, 285 MB unpacked, assembled by scripts/build_deposit.py, whose MANIFEST pairs each canonical result file with the script that regenerates it. sim81_deposit_v1.1.zip holds the simulation panel for the design-sensitivity analysis and is unchanged since v1.1. Main deposit. Codon alignments are restricted to the 35 published genomes among the 51 with coding sequence available; sequence from genomes that had no associated publication on the JGI MycoCosm portal at the time of analysis is not redistributed here. The trimmed set retains 9,620 sequences. Trees carry no sequence and cover all 51 tips. The deposit also carries per-genome enzyme counts, the 182-tip species tree, lifestyle assignments before and after an annotation audit, catalytic-residue calls, 38 canonical summary result files, seven figure audit tables recording every plotted value at full precision, and 170 analysis scripts. Two classes of artefact cannot be regenerated from a seed, and are deposited for that reason rather than for convenience: the null-set and internal-labelling trees, whose draws combined a seed constant with a per-process salted hash, and all 304 simulated alignments, whose generator passed no seed to the sequence simulator. Both defects are described in the manuscripts rather than worked around silently. Genome assemblies and predicted proteomes were obtained from the JGI MycoCosm portal and from NCBI and are not redistributed here. Related work. The material deposited here underlies two companion manuscripts. One documents that RELAX does not reproduce its own estimates at default settings and that its output does not report the failure; it is available as a preprint (doi:10.64898/2026.09.20.751138, posted 25 September 2026). The other, on how the placement of the test label determines the type I error of a branch-site test of relaxed selection, is in preparation. Both arise from an earlier preprint (doi:10.64898/2026.07.20.739482) whose relaxed-selection claims are withdrawn; the revised version withdrawing them was posted on 10 September 2026. Version history. v1.1 added the 78-gene panel replication with recorded seeds and its pre-registration. v1.3 adds the material behind the reproducibility manuscript — the seed-mechanism and random-number-dependence results, a twenty-gene four-arm reproducibility experiment with its pre-registration, and 529 HyPhy JSON and log files together with the seeded RELAX batch file, a minimal reproducible example and the run scripts — raising the deposit from 1,184 to 1,776 files. v1.4 removes a superseded copy of the main archive, aicomp-h1h3-deposit.zip from version 1.0, that had remained attached to the record alongside the current one, and updates this description. v1.5 corrects three ways in which the archive had drifted from the manuscripts. It adds the figure audit table for a correction figure drawn after v1.3; it refreshes the two figure scripts, whose deposited copies predated the submitted figures and could not draw them; and it makes the version file, the archive directory and the filename carry the number of the record. A README is added. No deposited result changed.