I'm confused with DNN predicts

[kezhouqin]

hi, Dr. Zhang,
I'm running the Darts_DNN, and I have several questions to ask. I have run Darts_BHT, and I have learned Darts_DNN get_data, and Darts_DNN build_feature, now I'am confused with Darts_DNN predict.

when I use Darts_BHT, I import the bam files obtained by star soft, but Darts_DNN build_feature requires kallisto, so I'am confused, whether I can use the rsem to obtain the expression data? could you tell me how I can get the file ENCODE_sequenceFeature_absmax_normalized.h5, and RBP_tpm.txt? is RBP_tpm.txt the expression file obtained by kallisto? I have learned some information from the issues, and will have a try. Another question is bam files which I used should be generated by star or by kallisto? Thank you very much!

Best wishes!

kezhou

[zj-zhang]

Hi to answer your questions:

1. You can use rsem for the RBP expression quantification, as long as the transcript mapping is the same as kallisto; however, it's quite possible that they are not.
   The current transcript to gene mapping is hard-coded which is far from extensible. Pull requests to fix this are welcomed - should make the expression quantification a lot more easy to use.

2. The files you mentioned are cis- and trans-features. They can be downloaded by
   Darts_DNN get_data -d transFeature cisFeature trainedParam -t ${eventType}
   I would recommend running the test data first:
   https://darts-dnn.readthedocs.io/en/latest/#using-predict

3. The bam files to run Darts_BHT should be generated by STAR, not kallisto.

Hope this helps.

[kezhouqin]

Thanks for you quick reply. It works on the test data.
other qestions: There is a parameter --readlength in Darts BHT, do you mean that it is seqence length (about 50bp) without primer adapter , not 150bp in RNA-seq? I remember Darts BHT depends on STAR soft, Darts DNN requires kallisto to further prove the AS event, is it right?

[zj-zhang]

I am not entirely sure what 50bp and 150bp means in your particular case, or why the sequencing adapter would take 100bp which looks odd to me. In any case, -t is the read length in BAM that your STAR is run against.
Right, Darts DNN requires the gene expression to improve the AS event.

[kezhouqin]

I am not entirely sure what 50bp and 150bp means in your particular case, or why the sequencing adapter would take 100bp which looks odd to me. In any case, -t is the read length in BAM that your STAR is run against.
Right, Darts DNN requires the gene expression to improve the AS event.

Oh，I‘m sorry, I mean that 50bp is the shortest length from the result tested by the soft fastqc, and 150bp is the length based on illumina second-generation sequencing platform.

The readlength of all sequences from the BAM file is different, isn't it? so how do I choose the length? Thanks!