How Can the OPLS-DA Permutation Test Plot Be Considered Not Overfitted
When applying the OPLS-DA (Orthogonal Partial Least Squares Discriminant Analysis) model for permutation testing, it is crucial to evaluate the model's predictive accuracy, particularly the Q2 statistic, in relation to the distribution of Q2 values obtained from the permutations. Specifically:
The Q2 statistic of the original model should exceed the distribution of Q2 values derived from the permuted datasets. This indicates that the model's performance with the original dataset is substantially better than with randomly permuted data. In the permutation test plot, the Q2 value of the original model should appear at the upper end of the distribution generated by permutations.
Permutation test p-value: This p-value is calculated based on how often randomly permuted models achieve a Q2 value that matches or exceeds that of the original model. A low p-value (typically below 0.05) suggests that the model’s predictive power is unlikely to be due to random chance, thereby diminishing the risk of overfitting.
Examination of the permutation plot: In the permutation test graph, the x-axis represents the permutation iterations, while the y-axis displays the corresponding Q2 values. The Q2 value from the original model should be clearly separated from the central tendency of the Q2 values from the permutations, indicating a shift toward higher values. If the original Q2 value overlaps significantly with those from the permutations, it signals a potential overfitting issue.
MtoZ Biolabs, an integrated chromatography and mass spectrometry (MS) services provider.
Related Services
How to order?
