Dear DeepMobilome authors,
Thank you for releasing the DeepMobilome code and for the very interesting work.
I am currently studying the paper and trying to understand the training data construction pipeline in detail, especially the implementation behind Table 1 and the Type1-Type7 cases shown in Fig. 2. I have a few questions related to reproducibility and would greatly appreciate any clarification.
-
In Table 1, the single-genome training set reports 33,773 target MGEs, including 6,526 positive target MGEs and 27,247 negative target MGEs. Could you clarify how these numbers were generated in the implementation?
Specifically, do the 33,773 target MGEs correspond to:
- unique natural MGE sequences collected before simulation,
- target/sample combinations generated during training data construction,
- or filtered alignment/SAM-level cases after read mapping?
-
Table 2 lists four genomes/strains used for single-genome training read simulation. Could you clarify their exact role in the code?
Were these genomes used as source/background genomes for simulating reads against a larger target MGE database, or were target MGE-related regions inserted, deleted, rearranged, or otherwise modified within these genomes to generate Type1-Type7 cases?
-
For the Type1-Type7 cases in Fig. 2, I would like to confirm how the labels were assigned in the implementation.
Were the Type labels predefined by constructing source genome/source sequence scenarios with known gene arrangements relative to each target MGE, followed by read simulation and alignment back to the target? Or were Type labels assigned after observing the read-to-target alignment patterns?
-
For negative cases such as Type2-Type6, were the rearrangements, deletions, and insertions generated at the gene level, for example moving/deleting/inserting complete MGE genes, or at nucleotide-level intervals?
-
If there are scripts or data-generation modules used to produce Table 1 and Type1-Type7 training cases that are not currently included in this repository, would it be possible to share them or provide a brief description/pseudocode of the implementation logic?
These details would be very helpful for reproducing the dataset construction process and correctly interpreting the reported training data counts.
Thank you very much for your time and for making DeepMobilome available.
Best regards
Dear DeepMobilome authors,
Thank you for releasing the DeepMobilome code and for the very interesting work.
I am currently studying the paper and trying to understand the training data construction pipeline in detail, especially the implementation behind Table 1 and the Type1-Type7 cases shown in Fig. 2. I have a few questions related to reproducibility and would greatly appreciate any clarification.
In Table 1, the single-genome training set reports 33,773 target MGEs, including 6,526 positive target MGEs and 27,247 negative target MGEs. Could you clarify how these numbers were generated in the implementation?
Specifically, do the 33,773 target MGEs correspond to:
Table 2 lists four genomes/strains used for single-genome training read simulation. Could you clarify their exact role in the code?
Were these genomes used as source/background genomes for simulating reads against a larger target MGE database, or were target MGE-related regions inserted, deleted, rearranged, or otherwise modified within these genomes to generate Type1-Type7 cases?
For the Type1-Type7 cases in Fig. 2, I would like to confirm how the labels were assigned in the implementation.
Were the Type labels predefined by constructing source genome/source sequence scenarios with known gene arrangements relative to each target MGE, followed by read simulation and alignment back to the target? Or were Type labels assigned after observing the read-to-target alignment patterns?
For negative cases such as Type2-Type6, were the rearrangements, deletions, and insertions generated at the gene level, for example moving/deleting/inserting complete MGE genes, or at nucleotide-level intervals?
If there are scripts or data-generation modules used to produce Table 1 and Type1-Type7 training cases that are not currently included in this repository, would it be possible to share them or provide a brief description/pseudocode of the implementation logic?
These details would be very helpful for reproducing the dataset construction process and correctly interpreting the reported training data counts.
Thank you very much for your time and for making DeepMobilome available.
Best regards