Does anyone have experience with Linear Discriminant Analysis (LDA) to distinguish mutant and control imaging-based phenotypes from each other? Can combinations of subtle differences lead to a biological meaningful distinction?
Lots of data is of course needed for LDA model development. How robust are those models when a large total number of images is present, but only a small number of images are compared directly i.e. in a high-throughput screening? For example a single drug-treated well (of which many exist) would be compared to a small number of DMSO treated wells.
Is anyone willing to share some LDA experiences concerning stability and test-ability of biological usefulness (for example validation of the LDA model with positive/negative control compounds)?
Thanks!
I recommend this paper : Article Hybrid random walk-linear discriminant analysis method for u...
it can be useful
Dear Johannes Wilbertz,
I used LDA models in my research papers to discriminate ex-mining ponds from lake water based on the selected/relevant water quality parameters. It worked perfectly with JMP software. I had good models and published many papers on this.
Though i'm not familiar with the biological terms you mentioned in your question, i believe you can use LDA to discriminate your samples based on the selected variables you mentioned. I will share some papers on LDA.
IMO It depends on the data and your goals. I recommend two books, full info can be found at these links: http://faculty.marshall.usc.edu/gareth-james/ISL/
and https://web.stanford.edu/~hastie/ElemStatLearn/
Please note that the newer methods with programs to be found in the text material are often preferable because they require less stringent assumptions on the data. Best wishes, David Booth
- Badges
- Science method