-
Notifications
You must be signed in to change notification settings - Fork 14
roadmap
Thomas Cokelaer edited this page Apr 18, 2016
·
12 revisions
Implemented
- Identical analysis as in the R version
- Regression analysis using OLS
- FDR correction using BH or qvalues method.
- HTML output as in R version
- All figures as in R version
What's new:
- Open cosmic browser in the HTML report
- Settings used are available in the HTML report
- HTML table with cosmic ids provided
- 2-3 times faster
- Descriptive analysis and visualisation of the IC50, genomic features
- Javascript on the volcano plot: we decided to use mpl3d for experimental purpose but note that the GDSC data is then exposed to a dedicated website with javascript embeded so we won't provide JS support officially.
- Include genomic data set.
- Standalone application
- Python notebooks
- Estimation of FDR using about 10 different methods
What could be included:
- Elastic Net: Implementation + test + documentation : 5 days
- Estimation of FDR with empirical pvalues: ?? 6-7 days ?? Not clear how to do that for now
- Training and support for developers to run gdsctools and generate data packages: 1-2 days
- Fully documented software: 5 days
- Include COSMIC
- Fetch IC50 data file on the web automatically if possible from the web interface: 1-2 days
- 12Nov2015: on v17 data set, analysis takes about 16 minutes on a mac or a dell 6520 (265 drugs, 677 features).
- Using multicore analysis (4 cores) takes about 9-10 minutes (10 Nov). Surprinsingly only a factor 2 improvmenents. We know that 4 cores does not mean 4x faster. However, note that this a fraction is due to the usage of the low_memory option (set to False), which means the analysis is slower by 20% for sure.
- 24Oct2015 on v17 data set, analysis takes about 18-20 minutes on 265 drugs, 677 features
- 12Oct2015 on v17 data set, analysis takes about 22-25 minutes on 265 drugs, 677 features