testing pretrined model #65
Comments
|
@bingdiao-zbw The dream_challenge data should give you similar performance as what we presented in the paper. It may be the case that you are using wrong alignments, or parameters. Would you please explain the details of your dream_challenge test case, the alignment, and the preprocess/call parameters you use? |
|
Thank you for your reply. The parameters are the same as mentioned in the run_test.sh. I found another problem when I run the preprocess.py. a lot of warnings appeared in the file scan.err. After running the call.py, The recall rate is less than 50% but the accuracy rate can reach more than 95%. |
|
@bingdiao-zbw In the paper, we used the following parameters for Dream data: |
|
|
Thank you for your reply. The information is the small region of the bam of the dream_challenge set5 |
|
@bingdiao-zbw thanks for sharing this. First of all, this is a different dataset compared to what we use in the paper. This one is Set5, but in the paper we used Dream challenge dataset for stages 3 and 4. But, regardless of this point, I don't expect low recall for this dataset. What I am suspecting now is that the aligner used for this dataset is not "BWA-MEM" (maybe "NovoAlign" or "BWA"). That can cause problem since we use some features that depend on the aligner. If that is the case you need to convert the bam to fastq and align again using BWA-MEM. Looking to mapping quality it seems to be that this is not aligned with BWA-MEM. And if you have downloaded the data from Dream Challenge website, I can see that they have used Novoalign v3.02.05 for stage 5 data (https://www.synapse.org/#!Synapse:syn312572/wiki/62018). So, you need to realign with BWA-MEM |
|
thank you very much. I will try according to your suggestion |
Hello, I tested my data and dream_challenge data with all the pre-trained model than you provided. The test results show that the recall rate is very low, less than 50%, I do not know the reason. can you provide me with the test data to verify the model, thank you
The text was updated successfully, but these errors were encountered: