Comments (18)
This is the schematics I showed before
from sccomp.
Hi Stefano,
I am still confused about how the one-step discovery rate is calculated. Is there a paradigm or algorithm for it?
from sccomp.
Hi Stefano,
I am still confused about how the one-step discovery rate is calculated. Is there a paradigm or algorithm for it?
Can you show me the results from an execution with the dev branch?
from sccomp.
from sccomp.
By one execution, do you refer to the sccomp_glm function?
Yes, can you paste the result table here?
from sccomp.
from sccomp.
The first excel is the data for regression, and the second one is table of result.
I don't see EXCELs here in github comments. You can simply copy and paste here the result table. I want to see what columns there are in your result table.
from sccomp.
cell_group .lower_(Intercept)
.lower_type1 .median_(Interce~ .median_type1
.upper_(Interc~ .upper_type1 significant
1 1 0.0514 0.181 0.430 0.761 0.841 1.36 TRUE
2 2 -0.481 -0.823 -0.0625 -0.264 0.423 0.327 FALSE
3 3 -0.479 -0.566 -0.0739 0.00293 0.396 0.619 FALSE
4 4 -0.588 -1.01 -0.156 -0.391 0.281 0.169 FALSE
5 5 -0.384 -0.813 0.00390 -0.139 0.452 0.418 FALSE
6 6 -0.452 -0.0177 -0.0679 0.577 0.393 1.15 FALSE
7 7 -0.340 -0.211 0.0538 0.372 0.488 0.968 FALSE
8 8 -0.688 -0.596 -0.271 0.000966 0.208 0.553 FALSE
9 9 -0.903 -0.744 -0.469 -0.173 0.00544 0.378 FALSE
10 10 -0.523 -1.14 -0.0574 -0.559 0.394 0.0581 FALSE
11 11 -0.00232 -0.595 0.398 0.00719 0.797 0.636 FALSE
12 12 -0.0513 -0.398 0.351 0.161 0.806 0.718 FALSE
13 13 -0.277 -0.0959 0.126 0.509 0.594 1.12 FALSE
14 14 -0.319 -0.812 0.114 -0.253 0.645 0.321 FALSE
15 15 -0.854 -0.862 -0.375 -0.290 0.155 0.270 FALSE
16 16 -0.437 -0.772 -0.0335 -0.182 0.410 0.396 FALSE
17 17 -0.378 -1.21 0.0478 -0.616 0.560 -0.0261 TRUE
18 18 -0.524 -0.495 -0.139 0.120 0.321 0.748 FALSE
19 19 0.100 -0.385 0.472 0.188 0.872 0.729 FALSE
20 20 -0.925 -0.402 -0.452 0.169 0.0590 0.754 FALSE
from sccomp.
If you install the dev version you should have false_positive_rate
from sccomp.
prob_non_zero false_discovery_rate concentration
1 0.0275 0.0275 <tibble [1 x 3]>
2 0.466 0.215 <tibble [1 x 3]>
3 0.990 0.470 <tibble [1 x 3]>
4 0.286 0.154 <tibble [1 x 3]>
5 0.702 0.388 <tibble [1 x 3]>
6 0.109 0.0732 <tibble [1 x 3]>
7 0.274 0.132 <tibble [1 x 3]>
8 0.999 0.496 <tibble [1 x 3]>
9 0.642 0.367 <tibble [1 x 3]>
10 0.142 0.0905 <tibble [1 x 3]>
11 0.987 0.441 <tibble [1 x 3]>
12 0.627 0.326 <tibble [1 x 3]>
13 0.158 0.104 <tibble [1 x 3]>
14 0.488 0.243 <tibble [1 x 3]>
15 0.392 0.184 <tibble [1 x 3]>
16 0.620 0.300 <tibble [1 x 3]>
17 0.083 0.0552 <tibble [1 x 3]>
18 0.74 0.409 <tibble [1 x 3]>
19 0.560 0.271 <tibble [1 x 3]>
20 0.632 0.347 <tibble [1 x 3]>
from sccomp.
Ok, did you learn everything about false discovery rate?
Do you know now how to test for how many false positives are in the first N positions of your false discovery rate rank?
from sccomp.
If you studied false discovery rate you will see that if you order the column according to that, the first N estimates, should have on average false_discovery_rate
probability of being false. Then you can test whether this is the case from N that goes from 1 to ...
from sccomp.
In this case, false discovery rate should be a non-decreasing function on (number of) category N ? N*false_discovery_rate == false_positive_rate?
from sccomp.
N*false_discovery_rate == average number of false discoveries.
For example
- if you have 20 categories with a false discovery rate < 0.5, on average you will have 10 categories that will be errors/false positives
- if you have 30 categories with a false discovery rate < 0.5, on average you will have 15 categories that will be errors/false positives
- if you have 30 categories with a false discovery rate < 0.1, on average you will have 3 categories that will be errors/false positives
from sccomp.
Then, what do I need to calculate or analysize?
from sccomp.
Same calibration you have done already.
Estimated false positives vs real false positive, for false_discovery_rate of 0.01, 0.02, 0.03, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5 ...
from sccomp.
the Estimated false positives should be N*false discovery rate, right?
from sccomp.
fp_aimed fp_estimated
0.00 0.000
0.01 0.000
0.02 0.015
0.03 0.040
0.05 0.070
0.10 0.165
0.20 0.370
0.30 0.570
0.40 0.785
0.50 0.940
0.60 1.000
0.70 1.000
0.80 1.000
0.90 1.000
1.00 1.000
from sccomp.
Related Issues (20)
- use of inv_logit vs softmax HOT 3
- throw error if missing values due to pivoting long HOT 3
- Best strategy for differential composition estimation in integrated Seurat object HOT 1
- stanfit correspondence parameters variable names HOT 18
- FDR calculation HOT 3
- Question about "test_composition_above_logit_fold_change" parameter in sccomp_glm HOT 3
- dirichlet_multinomial noise model doesn't seem to function HOT 4
- Interaction terms in sccomp_glm HOT 12
- Warings from sccomp_remove_outliers() when sample size is large HOT 1
- Does sccomp require replicates ? HOT 2
- Relationship between CI, minimal effect threshold, and FDR on plots$credible_intervals_1D HOT 2
- Fig 3B and 3 group comparisson HOT 1
- question: scope of the statistical model HOT 6
- sccomp_boxplot() errors when passing a factor variable HOT 3
- custom ggplot theme misses theme element HOT 1
- sccomp plot method -> rstan::gqs throws warning HOT 3
- question about lumping or selecting compositional data HOT 1
- Interpretation when both bars are highlighted versus one HOT 14
- Multilevel model error HOT 2
- Continuous variable as main effect HOT 4
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from sccomp.