Tumor Heterogeneity Analysis (THetA) and THetA2 are algorithms that estimate the tumor purity and clonal/subclonal copy number aberrations directly from high-throughput DNA sequencing data. This repository includes the updated algorithm, called THetA2.
Thank you for making this tool available! I am a bioinformatics engineer working with the Gabriella Miller Kids First Pediatric Research Program at the Children's Hospital of Philadelphia, and we use THetA2 in one of our workflows.
We recently noticed that the program was taking a while to finish with some large inputs, and after looking into it I believe that a considerable speedup can be achieved by vectorizing the array operations in the L2 and L3 functions of the CalcAllC module. I have implemented these changes in this pull request, and I would be grateful if you would be willing to discuss them.
Included in the PR are Python pickle objects containing inputs and outputs for the modified functions. L2args.pkl and L3args.pkl contain inputs for the functions on both the master branch and this one. The objects can be unpacked by e.g.
with open('L2args.pkl', 'rb') as pkl:
mu, C, m, r = pickle.load(pkl)
and likewise for L3, except that there is an additional n argument.
The other included pickle files contain the outputs of both functions on the master branch and this one. To unpack, for example, the outputs of L2 from this branch,
with open('L2outputs_branch.pkl', 'rb') as pkl:
total_sum, vals = pickle.load(pkl)
I am including these files to assist in comparing the outputs of the versions of the functions, which are the same to ~15 decimal places. My intention is that they would be removed prior to merge.
Python 2.7.15 was used for this work, as well as numpy 1.15.4.
One potential issue concerns the arguments m of L2 and m and n of L3. Although I did not modify the call signature of the functions, these arguments are not used in my proposed versions. I found them to refer to the dimensions of the array C. I added code to catch any instance in which their values differed from the dimensions of C, and I was able to run RunTHetA.py on our large inputs without these exceptions being raised. For this reason I believe that my understanding of the origin of the values of m and (in L3) n is correct, but I would be grateful if you were able to tell me that this relationship between the arguments cannot be assumed.
I appreciate your consideration of this request, and thank you again for making THetA2 available.
Dear ThetA2 team,
Thank you for making this tool available! I am a bioinformatics engineer working with the Gabriella Miller Kids First Pediatric Research Program at the Children's Hospital of Philadelphia, and we use THetA2 in one of our workflows.
We recently noticed that the program was taking a while to finish with some large inputs, and after looking into it I believe that a considerable speedup can be achieved by vectorizing the array operations in the
L2andL3functions of theCalcAllCmodule. I have implemented these changes in this pull request, and I would be grateful if you would be willing to discuss them.Included in the PR are Python
pickleobjects containing inputs and outputs for the modified functions.L2args.pklandL3args.pklcontain inputs for the functions on both the master branch and this one. The objects can be unpacked by e.g.and likewise for
L3, except that there is an additionalnargument.The other included
picklefiles contain the outputs of both functions on the master branch and this one. To unpack, for example, the outputs ofL2from this branch,I am including these files to assist in comparing the outputs of the versions of the functions, which are the same to ~15 decimal places. My intention is that they would be removed prior to merge.
Python 2.7.15 was used for this work, as well as numpy 1.15.4.
One potential issue concerns the arguments
mofL2andmandnofL3. Although I did not modify the call signature of the functions, these arguments are not used in my proposed versions. I found them to refer to the dimensions of the arrayC. I added code to catch any instance in which their values differed from the dimensions ofC, and I was able to runRunTHetA.pyon our large inputs without these exceptions being raised. For this reason I believe that my understanding of the origin of the values ofmand (inL3)nis correct, but I would be grateful if you were able to tell me that this relationship between the arguments cannot be assumed.I appreciate your consideration of this request, and thank you again for making THetA2 available.
Best, Nathan Johnson