Paper | Drug feature and protein feature | Method for negative samples | Description | Method |
---|---|---|---|---|
DTI-SNNFRA [30] (2021) | Drug: constitutional, topological, and geometrical descriptors. Protein: amino acid, pseudoamino acid, and CTD | First is the similarity between the drugs and the proteins. Then, the shared nearest neighbors and k-medoids clustering | First, the similarity between the drugs and the proteins. Then, the shared nearest neighbors and k-medoids clustered using the RUSBoost classifier for the prediction stage. | 1. Shared nearest neighbors 2. RUSBoost Classifier |
DeepCon [31] (2019) | Drug: Morgan fingerprint Protein: CNN on raw protein sequence, CTD | Dependent on the similarity between the drugs and the proteins; then compute the distance between the drug and protein. | First compute the distance depending on the similarity of drug and target features for predict the negative samples to achieved the class balance, second apply to DBN for prediction stage. | 1. The similarity of drug and target features 2. Deep belief network (DBN) |
Idti-MLKdr [32] (2021) | Drug: Morgan fingerprint Drug: AAC, DC, TC | evaluate the molecular similarity of drug and target features based on the Tanimoto coefficient (TC). Then, the Cluster-Based Molecular Similarity algorithm calculates and selects the top-ranked drugs and targets. | The Tanimoto coefficient (TC) depends on the similarity between the drugs and between the proteins. Then, use Cluster algorithm and finally using Multikernel learning (MKL). | 1. Cluster algorithm 2. Multikernel learning (MKL) |
PreDTIs [33] (2021) | Drug: drug-molecular substructure pattern fingerprint Protein: Psepssm | Using the SVM classifier. Then, the Euclidean distance is calculated from the predicted and the value of the real features | Use the SVM classifier. Then, calculate the Euclidean distance between the real and predicted values, using the LightGBM for prediction. | 1. Euclidean distance 2. LightGBM Classifier |
[20] (2020) | Drug: molecular substructure fingerprints Protein: Apply the PSSM, and then, apply the LOOP method to extract protein feature | Randomly select the number of negative samples, which is the same as the number of positive samples. | Randomly select the negative samples, equal to the positive samples. Apply the rotation forest for prediction. | 1. Rotation forest |
[35] (2020) | Drug: Morgan fingerprint. Protein: 20 amino acids | The negative sample sets consist of the same number of randomly selected pairs of unrelated drugs and proteins. | Randomly select the negative samples. Apply Random Forest for prediction. | 1. Random Forest classifier |
[16] (2017) | Drug: molecular descriptors and molecular fingerprints (MFs). Protein: AAC, DC, and TC | The negative dataset can be randomly selected from the DTS. | Random select the negative samples. Apply the deep belief network for prediction | 1. Deep belief network (DBN) |
[34] (2020) | Drug: (E-state) fingerprints Protein: (APAAC) | The Euclidean distance from all unlabeled samples to the positive center is calculated and sorted. The farther the distance is, the more likely the sample is to be negative. | The Euclidean distance from all unlabeled samples to the positive center. Apply support vector machines (SVM) for prediction. | 1. Euclidean distance 2. Support vector machines (SVM) |