dimiboeckaerts / bacteriophagehostprediction Goto Github PK

Good morning,

Could you share RBP_alignmentscores.txt file required to final preprocessing in RBP_host_prediction.ipynb?

Regards,
Piotr

Can you kindly include RBP_alignmentscores.txt in the repo so that I can run the final preprocessing step in the notebook that you provided? Thanks!

Hi,

I was using the RBP_functions.py module as a reference to extract features of certain proteins, but I noticed a possible bug. In the CTDC function, the loop overrides the encoding variable in each iteration, which is returned at the end. This is the code:

This causes to only consider the last of the properties (list defined in this line), which is not the desired behavior, I think. You can confirm this by replacing the whole list of properties with a list only containing the last value: ["solventaccess"]. I guess that the desired behavior is to calculate the encoding for each property and accumulate the whole sequence, resulting in a list of 39 values instead of only 3.

This possible error is also present in the supplementary material of the corresponding paper, as Table S3 (b) shows only 3 features for Protein composition, when it might be 39. In the reference paper (iFeature) the value is correctly described as 39 numerical features for the same purpose. The code of the reference paper does not have this possible bug and works as described:

https://github.com/Superzchen/iFeature/blob/942fb2760eae540a59214c540bd3329e8dca76e2/codes/CTDC.py#L75-L79

for p in property:
    c1 = Count(group1[p], sequence) / len(sequence)
    c2 = Count(group2[p], sequence) / len(sequence)
    c3 = 1 - c1 - c2
    code = code + [c1, c2, c3]

Is this actually a bug? If that's the case, does it affect the results displayed in the notebook significantly? I guess that more features might help to improve the models (depending on the nature of the missing features). Thanks.

Best,
Antonio.