possible data quality issues in dataset `0052`, specifically in: - 'raw_data/0052/0052_rtdata.tsv'

[keemwoojae]

The file appears to contain duplicate entries, missing structure annotations, and a likely misalignment between compound names/RTs and PubChem fields.

Observed Issues

1. Possible annotation shift in later rows

Rows 0052_00105-0052_00137 appear suspicious.
The compound name and rt fields seem to repeat earlier compounds, but the structure annotations appear to be shifted/misaligned.

Example:

• 0052_00105
  • name: Myo-inositol
  • rt: 1.1
  • formula: C22H18O11
  • PubChem CID: 65064
  • InChIKey: starts with WMBWRE...

These structure fields appear to correspond to (-)-Epigallocatechin gallate, not Myo-inositol.

Similar mismatches repeat through 0052_00137.

2. Duplicate-related comments are widespread

The comment column contains many duplicate-related notes:

• doublet: 62 rows
• removed another duplicate entry: 69 rows
• potential duplicate: 4 rows
• standardized from inchi; removed another duplicate entry: 1 row

It is unclear which entries are intended biological/analytical doublets and which should be removed or consolidated.

Please Check

Could you please check whether:

• rows 0052_00105-0052_00137 have shifted PubChem annotations,
• the doublet / potential duplicate / removed another duplicate entry labels reflect the intended final state,

Thanks for maintaining RepoRT. This dataset is very useful for retention time prediction benchmarking.

[f-kretschmer]

Thank you for spotting this! Have you by any chance already looked into the "original" PredRet dataset (CS14), if it has the same issue? I'll investigate this further

[keemwoojae]

I checked the original PredRet dataset/system at http://predret.org/retention-time-prediction/. (Please note that the "Retention time prediction" tab doesn't seem to be accessible via HTTPS.)

It appears that structural annotations like formula, SMILES, and CID are not provided there, but myo-inositol is indeed listed with an RT of 1.1.

[f-kretschmer]

I now looked into it more, I think you are right and all rows starting from 0052_00105 are faulty entries. Structures and retention times are exactly the same as the first rows. I think it is safe to assume that this was an error from copying/converting. I will delete the rows (of course you can also make a PR; I might not get to it very soon).

As for the doublets:

• removed another duplicate entry you can ignore (just use the data normally), it just means that there has been a duplicate entry that was already removed
• potential duplicate are entries with same structure and retention time which we are not entirely sure are duplicates, as e.g., the name might be different, and we just do not have enough structure information (stereochemistry missing). These have to be removed in evaluation to avoid data leakage
• doublet: same structure but different retention times, could be misannotations, uncertainty from dataset submitters, or missing (stereochemistry) information. Also have to be removed from evaluation, but can be used in model training.

After the faulty entries are removed, some of these comments will change, I'll update them.