biogenies/imputomics

imputomics's Issues

Imputation methods

1. min
2. mean
3. halfmin
4. median
5. zero
6. random
7. bpca
8. ppca
9. svd
10. nipals
11. nlpca
12. missmda_reg
13. missmda_em
14. amelia
15. missforest
16. mi
17. areg
18. qrilc
19. mle
20. twlsa
21. softimpute
22. irmi
23. pmm
24. midastouch
25. cart
26. rf
27. norm
28. norm.nob
29. norm.boot
30. norm.predict
31. lasso.norm
32. lasso.select.norm
33. PEMM https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4061266/
34. kNN-TN https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-017-1547-6
35. GSimp https://github.com/WandeRum/GSimp

Presentation synopsis

1. Aim: benchmark the quality of missing data imputation and find if imputing data is necessary for decision-making based on the metabolomics data.
   2. Scheme: https://viewer.diagrams.net/?page-id=_Gg_AfVaFpc6PrHJ8jsU&highlight=0000ff&edit=_blank&layers=1&nav=1&page-id=_Gg_AfVaFpc6PrHJ8jsU#G1z9yDcNV2doGwIQaU7gxAmNQtFy5LXXCc
2. Types of missing values.
3. Considered imputation methods.
4. "Safe imputation" https://github.com/michbur/metabolomics-missing-value-imputation/blob/main/functions/imputing.R#L68
5. Problems:
   7. How to measure the impact on modeling?
   8. Is there a single measure incorporating the quality of imputation and the speed of the imputation?

Benchmarks

• https://www.nature.com/articles/s41598-017-19120-0 and they have a tool: https://metabolomics.cc.hawaii.edu/software/MetImp/
• https://pubmed.ncbi.nlm.nih.gov/35881554/ and they have R package https://github.com/BeanLabASU/metabimpute
• 10.1007/s11306-018-1420-2

App specification

1. Input: single xlsx or csv file.
2. Tab: input data
3. Selection of the numeric-only columns.
4. Prepare 20 methods named methodX where X is a numeric constant. Each method imputes missing data by filling it with its constant. Eg., method11 fills every missing value with 11.
5. Provide a selection tab for selecting of the methods. Prepare also button "select all", "deselect all"
6. Prepare an output where each imputation method has its tab (or maybe if only less than 5 methods are chosen).
7. Prepare download buttons: download csv (all imputations in separate csv, file in a zip format) or xlsx where each sheet represents a different missing value imputation method.

Make sure that wrappers are correct

• make sure that scaling is applied before all wrappers
• make sure that wrappers are called correctly
• make sure that hyperparameters are estimated correctly

Methods of initialization

Some methods (e.g., GSimp) require initialization for missing values. Should we initialize all the methods?