Trader as a new optimization algorithm predicts drug-target interactions efficiently

Abstract

Several machine learning approaches have been proposed for predicting new benefits of the existing drugs. Although these methods have introduced new usage(s) of some medications, efficient methods can lead to more accurate predictions. To this end, we proposed a novel machine learning method which is based on a new optimization algorithm, named Trader. To show the capabilities of the proposed algorithm which can be applied to the different scope of science, it was compared with ten other state-of-the-art optimization algorithms based on the standard and advanced benchmark functions. Next, a multi-layer artificial neural network was designed and trained by Trader to predict drug-target interactions (DTIs). Finally, the functionality of the proposed method was investigated on some DTIs datasets and compared with other methods. The data obtained by Trader showed that it eliminates the disadvantages of different optimization algorithms, resulting in a better outcome. Further, the proposed machine learning method was found to achieve a significant level of performance compared to the other popular and efficient approaches in predicting unknown DTIs. All the implemented source codes are freely available at https://github.com/LBBSoft/Trader .

Figure 1

Pseudocodes for generating the dataset. The generated datasets only include positive drug-target interactions and have been obtained based on the chemical similarity score of drugs and smith waterman alignment score of targets.

Figure 2

The framework of the proposed method for drug repurposing. After generating the datasets, Trader trains the ANN using datasets. When the ANN is appropriately trained, the model is generated and then applied to the prediction of the unknown drug-target interactions. IN, H, D, and T show neurons of the input layer, and neurons of hidden layers, a drug, and a target, respectively.

Figure 3

The flowchart of Trader: The proposed optimization algorithm starts with some candidate solutions which each of them determine the weights of the ANN. Next, they are placed into several groups and are improved by Eq. 6 through 8 (see the text for details). The steps of Trader are repeated until the termination condition is satisfied. By passing the steps of the algorithm, the value of RMSE is also reduced and a suitable predictor model is acquired.

Figure 4

The pseudocode of Trader. For training the ANN, Trader produces some potential answers which consist of several variables (the edges of the ANN). Trader includes three operations, shown by Eq. 6 through 8. These operations change the weight of ANN’s edges differently. For instance, Eq. (7) alters them based on their content, or Eq. (8) tries to improve them by importing some values from the best solutions.

Figure 5

The convergence of the algorithms on different test functions shown by F. For instance, F_i presents i^th test function. (a) The average convergence of the algorithms on F1 through F9 and F15. (b) The average convergence of the algorithms on F11 and F12. (c) The convergence of the algorithms on F13. (d) The average convergence of the algorithms on F10, F14, and F16 through F20. Among the test functions, F10, F14, and F16 through F20 are the benchmark functions with the small sizes, but the others have a large number of variables with a higher range. These diagrams show that Trader has more stable behavior than the others on different benchmark functions whereas EPO, TGA, and ION fall into local optima for some of them as F11, F12, and F13. Also, the results state that the performance of the algorithms is almost the same when the size of a problem or the number of variables is small.

Figure 6

The ROC curve of the methods on the four gold-standard datasets. (a) The ROC curves of the algorithms on the enzyme dataset. (b) The ROC curves of the algorithms on the ion channel dataset. (c) The ROC curves of the algorithms on the G-protein dataset. (d) The ROC curves of the algorithms on the nuclear receptor dataset. Besides the four plots, there are also the values of the AUC. Except for the enzyme dataset, the proposed method has obtained better results than others. Furthermore, Trader’s average value of the AUC is higher than four others. ANNTR: Trader-based Artificial Neural network; RFDT: Rotation Forest-based Drug-Target predictor; RVM: Relevance Vector Machine; BAY: Bayesian ranking-based.

Figure 7

The PR curve of the methods on the gold-standard datasets. (a) The PR curves of the algorithms on the enzyme dataset. (b) The PR curves of the algorithms on the ion channel dataset. (c) The PR curves of the algorithms on the G-protein dataset. (d) The PR curves of the algorithms on the nuclear receptor dataset. The size of the positive and negative datasets is the same. The PR curves show the proper performance of the proposed method relative to the others. The average value of Trader’s AUS is also higher than them. ANNTR: Trader-based Artificial Neural network; RFDT: Rotation Forest-based Drug-Target predictor; RVM: Relevance Vector Machine; BAY: Bayesian ranking-based.

Figure 8

The convergence behavior of Trader on all the generated datasets in training of the ANN. (a) The Convergence of Trader on the EN dataset. (b) The Convergence of Trader on the IC dataset. (c) The Convergence of Trader on the GP dataset. (d) The Convergence of Trader on the NR dataset. The results relate to the best-obtained outcomes from 50 distinct executions. For all the datasets, Trader has led to an acceptable value of the RMSE.