Finding a compound that shifts disease-associated gene activity in the opposite direction can help narrow a search for treatment candidates. Add a docking image that places the compound inside a protein, and the supporting evidence can look stronger. I rebuilt this analysis around a prior question: does the judgment survive a change in the comparison, even when both calculations look plausible together?
What is a reversal being compared with?
A transcriptomic profile describes the activity of many genes together. Here, “reversal” means a change opposite to the pattern of increased and decreased gene activity associated with a disease. An opposite direction alone does not establish that a compound treats the disease. It also matters whether the change is common across other conditions and whether it stands out against suitable reference compounds or controls.
I set aside the early manuscript’s mock-analysis framing and rebuilt the work around real data and control comparisons. This meant tracing where a compound was classified as a candidate and checking that the stated rules matched the result tables, then aligning the analysis with the manuscript.
Questions that remain after a docking score
Molecular docking uses molecular and protein structures to calculate plausible binding positions and scores. It does not establish experimental binding or treatment efficacy. Alongside scores for disease-related targets, I examined what happened at comparison targets.
Checking that a calculation handles known examples and checking that a candidate distinguishes a particular target are different tasks. I connected the transcriptomic and docking stages without allowing a pass at one stage to replace the next comparison, and kept the conclusions within what each comparison could support.
Separate the parts that can be recalculated
I also prepared code that reads the preserved result tables and reapplies the decision rules. That work concerns the decision layer supported by the retained data. It is not a fresh execution of the whole process from the original gene-data queries through docking. Where original inputs or execution settings were missing, I did not describe them as reconstructed.
The progress record dated 3 September 2026 describes a transition between publication routes. This account covers the question and the rebuilding of the analysis; it does not turn internal negative results into a definitive claim that the compounds have no effect. Once there is a live version of record and an active DOI, I plan to explain the publishable numbers alongside the verification process.