178

■Bio-mathematics, Statistics and Nano-Technologies: Mosquito Control Strategies

Figure 9.3: 3D model of QSAR MLR1 equation that correlates Rindex with critical pressure

(CP) and calculated molar refractivity (CMR).

since this dependence is often very complex. Therefore, MLR modeling is more often used

in linear QSAR modeling than the ULR. The number of independent variables in MLR

models is limited according to the Topliss-Costello rule which says that the ratio between

the number of the samples and the number of independent variables should be greater than

five. It is worth stressing that in MLR modeling, multicollinearity is one of limiting factors

and must be examined. This is usually done based on the value of variance inflation factor

(VIF; if VIF > 5, the multicollinearity is significant). If multicollinearity is significant, it

is better to apply PCR or PLS regression methods where the multicollinearity is one of

the conditions for their application. The application of ridge regression (RR), PCR and

PLS methods in modeling of repellent activity of a set of compounds toward females of

A. aegypti mosquitos was demonstrated in the study by Natarajan et al. 2008. MLR and

various machine learning approaches in the modeling of repellent activity to A. aegypti of

a series of carboxamides was done by Doucet et al. 2019. In the study by De et al. 2018, the

MLR modeling, stepwise regression (SR) and PLS regression methods were successfully

applied in QSAR modeling of larvicidal activity of plant derived compounds against Zika

virus vector A. aegypti based on in silico molecular descriptors.

9.3.3

Non-linear chemometric regression modeling of repellence index

The modeling based on the application of artificial neural networks (ANNs) has be-

come one of the most used non-linear regression methods in chemometric modeling of bio-

logical activity (Kovaˇcevi´c et al. 2018b). The ANNs are the structures composed of densely

connected adaptive process elements. They have the ability to mimic the basic character-

istics of the human brain because each neural network consists of artificial neurons that

have a task to mimic biological neurons. The efficiency of this technique is reflected in

the ability to recognize complex relationships between input and output variables without