What is the difference between OPLSDA and PLSDA in metabolomics analysis? Can I only report OPLSDA result？
Many thanks
Hi,
I'm not an expert in metabolomics analysis. Please find the link below for the difference between OPLSDA and PLSDA
Article Multivariate Analysis in Metabolomics
I think you can add either OPLSDA or PLSDA results in your paper at your first submission, which your dataset matches best. Then wait for the review comments. If some reviewers are curious about the other results, you can list as suggested in point to point response letter.
Hope it works
Best
O-PLS-DA uses the same basic statistic but uses a mathematical filter to remove systematic variance in the dataset that is unrelated to the sample class (or Y dummy variable). For example, if you were looking at coronary heart disease then the orthogonal variance might relate to sex since men are more likely to develop the disease than women.
You can remove more than one orthogonal component, but the more orthogonal components you remove, the greater the chance of over-fitting your data.
The orthogonal matrix is retained ad so you ca explore the 'orthogonal' components easily to fully understand your data set.
The method was invented by Svante Wold and Johann Trygg ad the best article to describe the technique is theirs:
Article Orthogonal Projections to Latent Structures (O-PLS)
Wold S, Trygg J, J. Chemometrics 2002 16: 119-128
There is also a good blog by Umetrics explaining its use
https://blog.umetrics.com/explaining-differences-or-grouping-data-opls-da-vs-pca-data-analysis
The technique has been used mainly inn Food analysis and disease diagnosis e.g.
Pohjanen E, et al A multivariate screening strategy for investigating metabolic effects of strenuous physical exercise in human serum. J Proteome Res. 2007 Jun;6(6):2113-20. doi: 10.1021/pr070007g.
Gavaghan CL, et al. Physiological variation in metabolic phenotyping and functional genomic studies: use of orthogonal signal correction and PLS-DA. FEBS Lett. 2002 Oct 23;530(1-3):191-6. doi: 10.1016/s0014-5793(02)03476-2.
Zieliński L, et al Chemometrics as a tool of origin determination of Polish monofloral and multifloral honeys. J Agric Food Chem. 2014 Apr 2;62(13):2973-81. doi: 10.1021/jf4056715.
Worley B, Powers R. PCA as a practical indicator of OPLS-DA model
reliability. Curr Metabolomics. 2016;4(2):97-103. doi:
10.2174/2213235X04666160613122429.
Variations on OPLS-DA continue to improve modelling capability e.g.
Jonsson P, et al Constrained randomization and multivariate effect projections improve information extraction and biomarker pattern discovery in metabolomics studies involving dependent samples. Metabolomics. 2015;11(6):1667-1678. doi:
10.1007/s11306-015-0818-3.
See the following links
Article Multivariate paired data analysis: multilevel PLSDA versus OPLSDA
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2834771/#:~:text=The%20advantage%20of%20OPLSDA%20compared,to%20the%20first%20predictive%20component.&text=Furthermore%2C%20in%20OPLSDA%20only%20a,a%20two%2Dclass%20problem).https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4465187/
https://www.researchgate.net/publication/229575850_OPLS_discriminant_analysis_combining_the_strengths_of_PLS-DA_and_SIMCA_classification/figures?lo=1&utm_source=google&utm_medium=organic
Good luck
Dear Researcher, I would like to say in a comment three things:
a) A usual version of PLS-DA only is able to properly discriminate samples, instead of classify. Kindly see details in a relevant literature entitled: "Discriminant analysis is an inappropriate method of authentication", TrAC Trends in Analytical Chemistry 78 (2016) 17-22, doi: https://doi.org/10.1016/j.trac.2016.01.010;
b) The version mentioned in (a) is the so-called PLS-DA hard discriminant model under traditional version. However, soft classification can be performed by the new version of PLS-DA under quadratic discriminant analysis. Kindly see details in a relevant literature entitled: "Multiclass partial least squares discriminant analysis: Taking the right way—A critical tutorial", Journal of Chemometrics. 2018;32:e3030, doi: https://doi.org/10.1002/cem.3030;
c) Be very careful in making conclusions regarding classification on the base of the PLS scores (OPLS-DA) without proper validation with a relevant test. Kindly see details in a relevant literature entitled: "Some common misunderstandings in chemometrics", Journal of Chemometrics 24 (2010) 558‐564, doi: https://doi.org/10.1002/cem.1346.
Dear Elaine Holmes. What do you mean with "remove". Creating more orthogonal components? thank you very much for your prior explanation.
Regards,
Oscar
OPLS-DA (or OPLS) is the combination of the filter OSC (orthogonal signal correction) and PLS-DA (or PLS). OSC removes information from the X block (e.g., spectroscopic data) orthogonal to the Y block (dummy variables in PLS-DA, or dependent variables to be quantified in PLS). Care should be taken with overfitting, and no more than two factors should be removed.
My research is on chemmometrics applied to analytical chemistry, mainly in applications dealing with food authenticiy, forensic problems, etc. Thus, I have limited experience with metabolomics (and other omics). I have the impression that the use of OPLS-DA is routine in omics, and I am not sure this is always the best choice. As originally proposed by Svante Wold, OSC is recommend typically for models in which the variance accounted for the first components is much greater in X than in Y.
Here a nice reading:
Article Notes on the practical utility of OPLS
- Badges
- Science topic
- Similar topics
- Filing