I revisited public root RNA data associated with Panax, the plant genus that includes ginseng, to examine sequences resembling viral material. Even the word “candidate” needed a precise meaning. A sequence supported by the original reads, a rediscovery of a known sequence, and a sequence better explained by another organism require different interpretations.
In the report prepared on 19 August 2026, some earlier candidates were reassigned as internal controls corresponding to known viruses. The remaining sequences were limited to partial candidates, not complete genomes. A later automated-check report recorded successful source-data checks alongside an incomplete final database comparison. The work did not reach a claim of completed publication-level validation.
Checking for rediscovery
A search for unfamiliar sequences can recover material that has already been deposited. Alongside the observation that a protein region resembled viral material, I compared candidates against known sequences. NCBI RefSeq, used in that comparison, is a collection of reference genomic, transcript and protein sequences.
The 19 August report records a correction: some earlier candidates were near-identical to deposited Panax cryptic virus 1 and 2 sequences. I separated those entries from the novel-candidate list and treated them as positive controls, asking whether the analysis could recover known material. Their role changed from a possible finding to a check on the procedure.
The comparison also kept identity and alignment coverage separate. Identity describes the matching positions within an alignment; coverage describes how much of the query is included. A high match over a short segment supplies different evidence from a match across most of a sequence.
Read support has more than one dimension
RNA-seq data contain short reads rather than a single finished long sequence. After obtaining a candidate, the original reads can be mapped back to it to inspect which positions they support.
The report separated depth from breadth of support. Depth counts the reads overlapping a position. Breadth describes how much of the reference has read support. A pile of reads in one region does not establish support across the whole sequence, so I kept the reference coverage alongside average depth for each source dataset.
A separate de novo assembly check asked whether the reads could reconstruct the candidate through their overlaps. Mapping begins with a chosen reference; de novo assembly offers a different route through the same source data. Agreement between them can be useful, but it is not an experiment on a newly collected biological sample.
Protein structure and recurrence are separate questions
One clue examined in the candidates was a protein region associated with RNA-dependent RNA polymerase, or RdRP, an enzyme that copies RNA. The report diagrams the order of the relevant regions. That arrangement justified further examination without turning a partial sequence into a complete viral genome.
I also kept support in the source data separate from recurrence in Panax datasets from other studies. A signal associated with a plant dataset does not by itself identify the plant as the true host. Other organisms present in the sample and effects of data processing remain possible explanations to examine. The 19 August report distinguished those comparisons and retained unresolved work.
A sequence candidate, an actual infection and a cause of root rot therefore require different evidence. This search did not establish infection, disease causation or a risk to people.
Where the final comparison stopped
The later automated workflow separated source-data checks from the final sequence decision. Its completion report records passing mapping, independent-fragment and de novo checks across the libraries, but the final sequence-gate run is labelled cancelled. Within that run, the nonviral protein comparison is recorded as a failure because it did not return valid search statistics and positive-control results. Other comparisons were cancelled or lacked outputs.
An unfinished search cannot be read as a search that found no match. The final aggregation step detected the missing results and failed, leaving a technical publication NO-GO. This was not a biological rejection of the candidates. The database comparisons required for a publication claim were incomplete.
Returning to the work would begin with completing those failed or cancelled comparisons and confirming that the controls behave as intended. Recovering sequence ends, confirmation in separate biological samples and distinguishing the host would still remain. The decision at the time was to preserve the candidates while withholding a declaration of completed validation with the last checks unfinished.