Fig. 2
LASSO regression results and feature selection of db/db mice wounds. (A) Mean-squared error (MSE) of LASSO regression against the regularization parameter, \(\lambda\). For better visualization, the horizontal axis is scaled using the logarithm (Log(\(\lambda\))). The error bar denotes one standard deviation of MSE from tenfold Cross Validation. The dotted vertical line indicates \(\lambda\) at the minimum of MSE (\(\lambda =0.150\)). (B) LASSO solution path. The lines indicate the changes of \({\varvec{\beta}}\)’s with respect to \(\lambda\). In both (A) and (B), the numbers on the top of the plots indicate the number of non-zero \({\varvec{\beta}}\)’s, i.e. the number of relevant features selected by LASSO. (C) Feature Selection by LASSO of db/db mice wounds. 8 Proteins from db/db mice wound healing with non-zero β’s are selected by LASSO with λ = 0.041, i.e., λ at the minimum MSE. The proteins are ordered from left to right with respect to the absolute value of βj. The magnitude of the LASSO coefficient, βj, indicates the relative contribution to the wound healing, while the sign shows the direction. For example, MMP-2 has the strongest correlation with the % open wound with negative correlation. On the other hand, Leptin has the second strongest correlation but is positively correlated with % open wound. (69 proteins analyzed and n = 17 measurements for Day 3, n = 18 for Days 5 and 10, and n = 19 for Day 16).
