Dear Multi-Embed team,
Thank you very much for releasing this interesting work. I am particularly interested in the survival analysis experiments and have been trying to reproduce the reported results as faithfully as possible.
However, my reproduced results differ substantially from those reported in the paper. I suspect that this may be due to differences in my data preprocessing or data selection procedures, so I would like to kindly ask for some clarification regarding the preprocessing details.
I would greatly appreciate some clarification on two points:
- Gene expression preprocessing
The supplementary materials mention filtering genes without counts using edgeR, selecting approximately 1,500 highly variable genes (HVGs) for each cancer type, followed by log1p transformation and standard normalization.
Could you please clarify:
What gene expression values were used as the starting input?
How exactly were the ~1,500 HVGs selected?
What does "standard normalization" refer to (e.g., gene-wise z-score)?
Were HVG selection and normalization performed on the whole cancer cohort or separately within each cross-validation training fold?
If possible, a preprocessing script or minimal example would be extremely helpful.
- Pathology slide selection
Based on the slide counts in Supplementary_Table_1_Datasets.xlsx, I was wondering whether the survival analysis used Tissue Slides rather than only Diagnostic Slides. For PAAD, the reported slide count seems possibly consistent with including both.
Could you please clarify whether Diagnostic Slides, Tissue Slides, or both were used, especially for PAAD?
In addition, were there any additional slide-level filtering or selection procedures before the survival analysis, such as filtering by sample type, slide quality, tumor content, or other criteria?
If a patient had multiple eligible slides after filtering, were all of them retained?
Thank you very much for your time and for making Multi-Embed publicly available. Any clarification would be very helpful for reproducing the survival analysis experiments.
Dear Multi-Embed team,
Thank you very much for releasing this interesting work. I am particularly interested in the survival analysis experiments and have been trying to reproduce the reported results as faithfully as possible.
However, my reproduced results differ substantially from those reported in the paper. I suspect that this may be due to differences in my data preprocessing or data selection procedures, so I would like to kindly ask for some clarification regarding the preprocessing details.
I would greatly appreciate some clarification on two points:
The supplementary materials mention filtering genes without counts using edgeR, selecting approximately 1,500 highly variable genes (HVGs) for each cancer type, followed by log1p transformation and standard normalization.
Could you please clarify:
What gene expression values were used as the starting input?
How exactly were the ~1,500 HVGs selected?
What does "standard normalization" refer to (e.g., gene-wise z-score)?
Were HVG selection and normalization performed on the whole cancer cohort or separately within each cross-validation training fold?
If possible, a preprocessing script or minimal example would be extremely helpful.
Based on the slide counts in
Supplementary_Table_1_Datasets.xlsx, I was wondering whether the survival analysis used Tissue Slides rather than only Diagnostic Slides. For PAAD, the reported slide count seems possibly consistent with including both.Could you please clarify whether Diagnostic Slides, Tissue Slides, or both were used, especially for PAAD?
In addition, were there any additional slide-level filtering or selection procedures before the survival analysis, such as filtering by sample type, slide quality, tumor content, or other criteria?
If a patient had multiple eligible slides after filtering, were all of them retained?
Thank you very much for your time and for making Multi-Embed publicly available. Any clarification would be very helpful for reproducing the survival analysis experiments.