Viral infectious diseases have been seriously threatening human health. The receptor binding is the first step of viral infection. Predicting virus–receptor interactions will be helpful for the interaction mechanism of viruses and receptors, and further find some effective ways of preventing and treating viral infectious diseases so as to reduce the morbidity and mortality caused by viruses. Some computation algorithms have been proposed for identifying potential virus–receptor interactions. However, a common problem in those methods is the presence of noise in the similarity network. A new computational model (Network Enhancement and the Regularized Least Squares [NERLS]) is proposed to predict virus–receptor interactions based on improving similarities by Network Enhancement (NE). NERLS integrates the virus sequence similarity, the receptor sequence similarity and known virus–receptor interactions. We compute the virus sequence similarity and known virus–receptor interactions to construct the virus similarity network. The receptor similarity network is constructed by the Gaussian interaction profile kernel similarity and the receptor sequence similarity. To obtain the final virus similarity network and the final receptor similarity network, NE is, respectively, applied for reducing the noise of the virus similarity network and the receptor similarity network. Finally, NERLS employs the regularized least squares to predict interactions of viruses and receptors. The experiment results show that NERLS achieves the area under curve value of 0.893 and 0.921 in 10-fold cross-validation and leave-one-out cross-validation, respectively, which is consistently superior to four related methods [which include Initial interaction scores method via the neighbors and the Laplacian regularized Least Square (IILLS), Bi-random walk on a heterogeneous network (BRWH), Laplacian regularized least squares classifier (LapRLS), and Collaborative matrix factorization (CMF)]. Furthermore, a case study also demonstrates that NERLS effectively predicts potential virus–receptor interactions.

中文翻译：

---

基于改进相似性的病毒-受体相互作用预测

病毒性传染病已严重威胁人类健康。受体结合是病毒感染的第一步。预测病毒-受体相互作用将有助于病毒与受体的相互作用机制，进一步寻找预防和治疗病毒性传染病的有效途径，从而降低病毒引起的发病率和死亡率。已经提出了一些计算算法来识别潜在的病毒-受体相互作用。然而，这些方法中的一个常见问题是相似性网络中存在噪声。提出了一种新的计算模型（网络增强和正则化最小二乘法 [NERLS]），基于通过网络增强 (NE) 提高相似性来预测病毒-受体相互作用。NERLS整合了病毒序列相似性，受体序列相似性和已知的病毒-受体相互作用。我们计算病毒序列相似性和已知病毒-受体相互作用来构建病毒相似性网络。受体相似性网络由高斯相互作用轮廓核相似性和受体序列相似性构成。为了得到最终的病毒相似性网络和最终的受体相似性网络，NE分别用于降低病毒相似性网络和受体相似性网络的噪声。最后，NERLS 使用正则化最小二乘来预测病毒和受体的相互作用。实验结果表明，NERLS在10倍交叉验证和留一法交叉验证中分别达到了0.893和0.921的曲线下面积值，这始终优于四种相关方法[包括通过邻居和拉普拉斯正则化最小二乘法 (IILLS) 的初始交互评分法、异构网络上的双随机游走 (BRWH)、拉普拉斯正则化最小二乘分类器 (LapRLS) 和协作矩阵分解（CMF）]。此外，一个案例研究还表明，NERLS 有效地预测了潜在的病毒-受体相互作用。