Related methods for evaluation

Nowadays, there are many usual methods in predicting the drug permeability of BBB, such as multi-core SVM, KNN, DT and so on. Therefore, we select several methods to compare with the Deep Learning method, which proposed in this paper in order to evaluate the performance of our method.

Multi-core SVM method is one of the most common methods in the published BBB permeability papers. For example, Gao et al. adopted POLY-SVM, RBF-SVM and normalized POLY-SVM methods in predicting the drug permeability of BBB²⁹.

The SVM method assumes the hyperplane equation is wT+b=0. Let x be a vector of N dimensional input space. Let ∅(x)=(ϕ1(x),ϕ2(x),...,ϕM(x))denote the nonlinear transformation from the input space to the M-dimensional feature space. A superclass plane can be constructed in this feature space and the equation is⁴⁸:

where wj is the weight that connects the feature space to the output space, and b is the offset.

If the data is not linearly separable, the kernel function will be used. The common kernel functions include Linear, Poly, RBF, Sigmoid and so on. Gao et al. paper proposed to use POLY-SVM, RBF-SVM and normalized POLY-SVM method in predicting the drug permeability of BBB which is based on clinical features²⁹. However, in normalized POLY-SVM, the normalization only uses to preprocess the data and its influence on the results is slight. Therefore, we use another high performing method named Sigmoid-SVM method instead of normalized POLY-SVM in comparison.

KNN method is a kind of the classical data mining methods and it also has been used to predict drug penetration of BBB in many years.

KNN method is measuring the distance between different feature values. Its main idea is that if a sample in the feature space, most similar samples of K, which means the nearest neighbors in the feature space, belong to a certain category, then the sample also belongs to this category, where K is usually not greater than an integer of 20²⁵.

Decision Tree (DT) looks like the tree structure, which can be a binary tree or a non-binary tree. Each non-leaf node represents a feature attribute, each branch represents the output of the feature attribute in a range of values, and each leaf node stores a category²⁷.

DT begins at the root node, then judge the corresponding feature in the item to be classified and selects the output branch according to its value until it reaches the leaf node. Finally, DT saved the category at the leaf node as the result of the decision⁴⁹.