Description:
Regression model with 5 QuBiLS-MIDAS descriptors used for the logarithmic values prediction of the Percentage of Repellency, PR, (%) to trigger Periplaneta americana cockroach repellency.

The training was performed with the Stacking algorithm as default in Weka 3.9.4 with 10-fold cross-validation by using Random Forest algorithm as meta classifier and as classifiers: Gaussian Processes (with Pearson Universal Kernel (PUK)), Linear Regression, and M5P. The 5 QuBiLS-MIDAS descriptors are namely:

AC[2]_K_Tr_AB_nCi_3_M20(M1)_SS4_T_KA_psa-a-c_MID
N1_TrC_AB_nCi_3_M21(M8)_SS7_T_LG3P[2]_LGP[2]_h_MID
AC[3]_K_Tr_AB_nCi_3_M20(M1)_NS6_T_KA_m-psa-c_MID
VC_Tr_AB_nCi_3_M21(M14)_NS4_T_LGBA[0.314-0.628]_psa-e-v_MID
I50_TrF_AB_nCi_3_M26(M15)_SS1_T_LGBA[0.314-0.628]_p_MID

Training set:
34 compounds extracted from 10.1002/cbdv.200890058

Performance:
For a 10-fold cross-validation, the statistical parameters (performance without applicability domain) are R = 0.8871, MAE = 5.4479, RMSE = 7.1032, RAE = 41.2105 %, and RRSE = 44.749 %.

Classification Breakpoint:
The breakpoint is 90 %. Values greater than or equal to the breakpoint will elicit a repellent response in Blatella germanica and Periplaneta americana cockroaches. Lower values represent certain actions occurring, however, these are not enough to activate a repellent reaction.

Reference:
Gaudin et al. Carboxamides Combining Favorable Olfactory Properties with Insect Repellency. 2008, 5(4), 617-635. DOI: 10.1002/cbdv.200890058