Placing small molecules around a protein in a computer model and scoring the resulting arrangements can help narrow a set of candidates. But if one calculation searches one region and another searches somewhere else, it is hard to tell whether a score difference reflects the protein states being compared or simply the search locations. I rebuilt the calculations to compare the region where OPTN and TBK1 meet under matched conditions.
The search for these positions and orientations is called molecular docking. It produces arrangements and scores under a chosen structure and set of computational assumptions. A convincing-looking fit on screen does not establish real binding or a treatment effect.
Matching the search space and repeats
An earlier exploration had compared protein states using different search-space centers. I could not recover the exact final inputs and output coordinates from that work, so I regenerated inputs from public structural records and carried out a separate calculation rather than claiming an exact reproduction.
I considered the wild-type, or WT, protein and the E50K variant in two structural chain pairs, A/C and B/D. That gives four receptor systems. Within each chain pair, WT and E50K used search boxes with the same position and dimensions. The public release calls this fixed-space design a controlled grid.
Eleven compounds were evaluated ten times in each of the four systems, producing 440 run records. The ten repeats use different random starting seeds; they are computational repeats, not ten biological experiments. Each run's inputs, commands, logs and output coordinates are included in the release, allowing someone to trace a calculation beyond its representative image. Public code and data, v2.0.0
Reading scores and repeatability separately
A molecule with a favorable score can still land somewhere very different when the seed changes. I therefore compared the positions produced by repeated searches separately from the scores. RMSD summarizes differences between the molecule's atomic positions in the same protein coordinate frame. Here, the calculation did not superimpose the molecules or exchange symmetry-equivalent atoms, so those choices belong with the interpretation.
The runs used Uni-Dock 1.1.3 with Vina scoring, fast search mode and 20 requested output arrangements per run. No explicit exhaustiveness argument was supplied. Each search box was a cube with 26 Å sides; the exact centers and input-preparation commands are recorded in the public package. An ångström is a unit used for distances at the atomic scale. These settings and coordinates provide a starting point for checking the calculation, without guaranteeing that the chosen protein structure or scoring function fully represents reality.
I wanted to leave a dataset whose comparison conditions and coordinates could be followed, rather than a ranking that declared a treatment candidate. Even recurring arrangements do not demonstrate physical binding. The calculations do not establish modulation of the OPTN–TBK1 interaction or an effect on glaucoma; those judgments require separate binding and functional experiments.
Comparing how much of the protein structure was included
I then examined the protein assembly included in the calculation, comparing inputs that retained two chains (pair-only) with inputs that retained four (four-chain). A chain is a connected sequence of amino acids within a protein structure. To interpret where a molecule lands, I need to consider how much of the surrounding structure was included, as well as the region where OPTN and TBK1 meet.
The new supplement contains 440 two-chain and 440 four-chain docking outputs calculated during the same period, together with the prepared four-chain structures, analysis results and scripts for replaying the analyses. This assembly comparison was absent from the August release. Its two sets of outputs should be distinguished from the 440 runs described earlier in this note.
I published the follow-up computational supplement on 29 September 2026 as nine database files and a Python extractor. With Python 3.9 or later, the extractor verifies the parts and restores a package of 3,033 files. The accompanying README provides the commands and SHA-256 checksums.
The code and data v2.0.0 released on 28 August remain available as a record of the earlier calculations. The two DOIs identify their respective computational datasets, not a published journal article.
I wrote up the comparison of two-chain and four-chain inputs, and the paper was published in Perspectives on Integrative Medicine on 20 October 2026. The published article (2026;5(3):226–233) ↗ describes the methods and results; the public datasets linked above provide the calculation conditions and coordinates.