NRLMF β: Beta-distribution-rescored neighborhood regularized logistic matrix factorization for improving the performance of drug-target interaction prediction

Abstract

Techniques for predicting interactions between a drug and a target (protein) are useful for strategic drug repositioning. Neighborhood regularized logistic matrix factorization (NRLMF) is one of the state-of-the-art drug-target interaction prediction methods; it is based on a statistical model using the Bernoulli distribution. However, the prediction is not accurate when drug-target interaction pairs have less interaction information (e.g., the sum of the number of ligands for a target and the number of target proteins for a drug). This study aimed to address this issue by proposing NRLMF with beta distribution rescoring (NRLMFβ), which is an algorithm to improve the score of NRLMF. The score of NRLMFβ is equivalent to the value of the original NRLMF score when the concentration of the beta distribution becomes infinity. The beta distribution is known as a conjugative prior distribution of the Bernoulli distribution and can reflect the amount of interaction information to its shape based on Bayesian inference. Therefore, in NRLMFβ, the beta distribution was used for rescoring the NRLMF score. In the evaluation experiment, we measured the average values of area under the receiver operating characteristics and area under precision versus recall and the 95% confidence intervals. The performance of NRLMFβ was found to be better than that of NRLMF in the four types of benchmark datasets. Thus, we concluded that NRLMFβ improved the prediction accuracy of NRLMF. The source code is available at https://github.com/akiyamalab/NRLMFb.

Keywords: Bayesian inference; Bayesian optimization; Beta distribution; Drug–target interaction prediction; Neighborhood regularized logistic matrix factorization; Rescoring.

Graphical abstract

Figure 1

A plot showing the probability density function of the beta distribution and the improved value of the score. (a) Shows the change in the score when the original score s (assuming NRLMF score) is less than 0.5. $s{\text{'}}_1$ is the concentration defined by $a+b$ (assuming the new feature γ) value, and $s{\text{'}}_2$ represents the score when the concentration value is small. Similarly, (b) shows the change in the score when the original score s is greater than 0.5.

Figure 2

Heatmaps of AUPR when hyperparameters ${\eta }_1,{\eta }_2$ are changed in each pair of cross-validation scenario (CVS) and dataset. The AUPR of each heatmap uses the average value of AUPR calculated for each CVS. However, hyperparameters without ${\eta }_1,{\eta }_2$ were fixed using different values for each pair of CVS and dataset. In addition, the frame in each figure is the search range of the hyperparameters ${\eta }_1,{\eta }_2$ defined in section 2.6.

Figure 3

Scatter plot of scores $s,s\text{'}$ and feature quantities γ of NRLMF and NRLMFβ. The score was calculated using the leave-one-out method corresponding to CVS3 for the nuclear receptor dataset. The horizontal axis represents the score of each method, and the vertical axis represents the feature quantity we introduced.

Figure 4

Histogram for NRLMF and NRLMFβ scores $s,s\text{'}$. The score was calculated using the leave-one-out method corresponding to CVS3 for the nuclear receptor dataset. The horizontal axis represents the score of each method, and the vertical axis represents frequency (cutoff = 50).