 Research
 Open Access
 Published:
Laplacian normalization and birandom walks on heterogeneous networks for predicting lncRNAdisease associations
BMC Systems Biology volumeÂ 12, ArticleÂ number:Â 122 (2018)
Abstract
Background
Evidences have increasingly indicated that lncRNAs (long noncoding RNAs) are deeply involved in important biological regulation processes leading to various human complex diseases. Experimental investigations of these disease associated lncRNAs are slow with high costs. Computational methods to infer potential associations between lncRNAs and diseases have become an effective priorpinpointing approach to the experimental verification.
Results
In this study, we develop a novel method for the prediction of lncRNAdisease associations using birandom walks on a network merging the similarities of lncRNAs and diseases. Particularly, this method applies a Laplacian technique to normalize the lncRNA similarity matrix and the disease similarity matrix before the construction of the lncRNA similarity network and disease similarity network. The two networks are then connected via existing lncRNAdisease associations. After that, birandom walks are applied on the heterogeneous network to predict the potential associations between the lncRNAs and the diseases. Experimental results demonstrate that the performance of our method is highly comparable to or better than the stateoftheart methods for predicting lncRNAdisease associations. Our analyses on three cancer data sets (breast cancer, lung cancer, and liver cancer) also indicate the usefulness of our method in practical applications.
Conclusions
Our proposed method, including the construction of the lncRNA similarity network and disease similarity network and the birandom walks algorithm on the heterogeneous network, could be used for prediction of potential associations between the lncRNAs and the diseases.
Background
Long noncoding RNAs (lncRNAs) form a new class of important ncRNAs, with length longer than 200nt [1â€“3]. Accumulating evidences have indicated that a large quantity of lncRNAs play critical roles in many important biological processes such as chromatin modification, transcriptional and posttranscriptional regulation, genomic splicing, differentiation, immune responses, and cell cycle control [1â€“4]. Mutations and dysregulations of these lncRNAs have been found to be linked to the development and progression of various complex human diseases [2, 3].
Computational models have been developed to predict potential associations between lncRNAs and diseases. Chen et al. [4] had an assumption that functionally similar lncRNAs tend to associate with similar diseases and vice versa. Based on this assumption, Chen et al. [4] proposed a method of Laplacian regularized least squares for lncRNAdisease association (LRLSLDA) to infer human lncRNAdisease associations. LRLSLDA calculates the Gaussian interaction profile kernel similarity for both diseases and lncRNAs based on known lncRNAdisease associations, and computes the lncRNA expression similarity of Spearman correlation coefficient between each lncRNA pair, then utilizes Laplacian regularized least squares in the lncRNA space and disease space to combine the optimal classifiers in these spaces to identify potential associations. LRLALDA is a semisupervised classification algorithm that does not require negative training samples. However, a major issue of LRLSLDA is how to combine two classifiers and how to select suitable parameters. Chen et al. [5] developed two novel calculation models for lncRNA functional similarity (LNCSIM). Chen et al. [6] proposed a fuzzy measurebased LNCRNA functional similarity computational model(FMLNCSIM). Chen et al. [7] introduced the model of KATZ measure to predict potential lncRNAdisease association.
Based on the fact that noncoding genes are often cooperated in human diseases to predict potential lncRNAdisease association, Peng et al. [8] proposed a new vector to represent diseases, and applied the newly vectorized data for a positive unlabeled learning algorithm to predict and rank diseaserelated lncRNAs. Ding et al. [9] proposed a model constructing lncRNAdiseasegene tripartite graphs (TPGLDA) includes genedisease associations and lncRNAdisease associations, and then applied to the process of resourceallocation on tripartite graphs to construct the potential lncRNAdisease association. However, TPGLDA only focuses on unweighted tripartite graphs.
Some models predict novel associations without referring to known associations between lncRNAs and diseases. Chen [10] proposed a model of hypergeometric distribution for lncRNAdisease association (HGLDA) to predict potential lncRNA disease associations. Zhou et al. [11] proposed a rankbased method called RWRHLD, which integrates the miRNAlncRNA association network, diseasedisease similarity network and known lncRNAdisease association network into a heterogeneous network and implementing a random walk with restart on this heterogeneous network to predict novel lncRNAdisease associations. However, RWRHLDA cannot be applied to lncRNAs without a known miRNA interaction partner.
Some computational models have been applied to predict lncRNAdisease associations based on random walk on networks. Chen et al. [12] considered the limitations of traditional random walks with restart (RWR), and proposed a model of improved random walk with restart (IRWRLDA) to predict lncRNAdisease associations. Sun et al. [13] proposed a method of RWRlncD based on global network to predict potential lncRNA disease associations. However, RWRlncD only considers lncRNAs which have known associations with the disease and ignores lncRNAs that are currently not associated with the disease. Considering the differences in the network topology of lncRNA and disease, Gu et al. [14] proposed a random walk model on global networks for predicting lncRNAdisease associations (GrwLDA). Yu et al. [15] proposed a model that performs birandom walks to predict lncRNAdisease associations (BRWLDA). However, BRWLDA only considers the semantic similarity of the disease, and the transitional probability between diseases is only empirically estimated.
In this study, we propose a novel computational model of Laplacian normalization and birandom walks on heterogeneous networks for predicting lncRNAdisease associations (LapBiRWRHLDA). Firstly, the method calculates the Gaussian interaction profile kernel similarity of lncRNAs and diseases by known lncRNAdisease associations. Next, we integrate the two sources of similarity to construct an lncRNAlncRNA similarity network. The diseasedisease similarity network can be constructed by the profile kernel similarity of diseases. Subsequently we perform Laplacian normalization on the similarity matrices of lncRNAs and diseases as the transpose matrices. Furthermore, we apply random walks on the lncRNA similarity network and the disease similarity network, respectively. Finally, we use a weighted average of random walks on both networks as a predictor of lncRNA disease associations. We believe that the higher scores of lncRNAdisease associations will have greater possibility for further verification. To evaluate our proposed method, we utilize leaveoneout crossvalidation experiments to demonstrate its superior performance compared with existing approaches. Furthermore, the analyses of three cancers (namely, breast cancer, lung cancer, and liver cancer) effectively support the practical application of our method. We then use LapBiRWRHLDA to infer potential lncRNAdisease associations. Some highscore results are successfully verified by the LncRNADisease and Lnc2Cancer databases.
Results
Leaveoneout crossvalidation
To assess the performance of our proposed method, we use the leaveoneout crossvalidation to perform the assessment. We leave out each known lncRNAdisease association in turn as test sample, while other known relationships are used as training samples and all unknown relationships are taken as candidate samples. Since disease similarity and lncRNA similarity depend on the Gaussian interaction profile kernel similarity of the known lncRNAdisease association, the disease similarity and lncRNA similarity will change when we delete a known lncRNAdisease association, so we will get different similarities.
A receiveroperating characteristics (ROC) curve is applied to determine the predictive performance, which plots the correlation between truepositive rate (TPR) indicating sensitivity and falsepositive rate (FPR) indicating specificity at different thresholds. Sensitivity represents the percentage of the leftout associations achieving the ranking higher than a given threshold; specificity means the percentage of candidate associations achieving the ranking lower than this given threshold. When we vary thresholds, we will obtain the corresponding different TPRs and FPRs. In this way, ROC is drawn and AUC is calculated. As a result, LapBiRWRHLDA achieved the AUC of 0.8409, 0.8527 and 0.8429 for three datasets used, respectively.
The effect of parameters in LapBiRWRHLDA
Parameter Î± controls the probability of the random walk restart. To optimize the parameter Î±, we increased Î± from 0.1 to 1 with step size 0.1, and then calculated the corresponding AUC value by LOOCV. After experimental verification, we chose Î±=0.9, and we achieved the AUC values of 0.8409. The experimental results indicate that LapBiRWRHLDA offers better performance on the LncRNADisease dataset on October 2012, when Î±=0.9 is selected. Similarly we achieved the AUC values of 0.8527 (Î±=0.2) and 0.8429 (Î±=0.8) based on Lnc2Cancer dataset on July 2016, and the LncRNADisease dataset on April 2016.
Performance comparison with other methods
We compared LapBiRWRHLDA with previous published methods in LOOCV based on the LncRNADisease dataset on October 2012. (1) LRLSLDA [4] computes Gaussian interaction profile kernel similarity for both diseases and lncRNAs from known lncRNAdisease associations and lncRNA expression profiles, and then applies the framework of Laplacian regularized least squares to identify potential associations. (2) GrwLDA [14] predicts potential associations by a random walk model on global networks for predicting lncRNAdisease associations (GrwLDA). The comparison is shown in Fig. 1. We also compared our method, LRLSLDA and GrwLDA in LOOCV based on the LncRNADisease dataset on April 2016. The comparison is shown in Fig. 2. Figure 3 shows the comparison of LapBiRWRHLDA, LRLSLDA and GrwLDA in LOOCV based on the Lnc2Cancer dataset on July 2016. These comparisons consistently indicate a better performance of our method over the stateoftheart methods for predicting lncRNAdisease associations.
We also compared the LncRNADisease dataset on April 2018 with the LncRNADisease dataset on October 2012, then selected 50 lncRNAdisease associations which were unverified in the LncRNADisease dataset on October 2012 but were verified in the LncRNADisease dataset on April 2018. We compared our method with LRLSLDA, GrwLDA by independently testing the ranking of the 50 relationships. Through experimental tests, our method has 30 rankings higher than the LRLSLDA method, and 41 rankings higher than the GrwLDA method.
Case studies
To further highlight the performance of LapBiRWRHLDA, we studied the predictive performance of three cancers: breast cancer, lung cancer, and liver cancer. For each type of cancer, we take the top 10 most probable lncRNAs as candidates associated with this cancer. Next, we manually checked these lncRNAs by mining biomedical literature from the LncRNADisease dataset and the Lnc2Cancer dataset.
Breast cancer is the second leading cause of female cancer deaths, comprising 22% of all cancers in women [16, 17]. LapBiRWRHLDA identifies potential lncRNAs associated with breast cancer and six of the top 10 verified by the recent LncRNADisease dataset. The list in Table 1 shows the lncRNAs associated with breast cancer. Lung cancer is one of the fastest increases in morbidity and mortality, and one of the greatest threats to human health and life. In the past 50 years, many countries have reported that the incidence and mortality of lung cancer are significantly higher. The incidence and mortality of male lung cancer account for the first place among all malignant tumors. LapBiRWRHLDA identifies eight out of the top 10 verified (see Table 2). Liver cancer is the fifth most commonly diagnosed cancer and the second most frequent cause of cancer deaths in men worldwide [18, 19]. LapBiRWHLDA correctly identifies five liver cancer related lncRNAs. Table 3 lists the lncRNAs related to liver cancer. From these case studies, we can conclude that LapBiRWRHLDA is a powerful tool for predicting lncRNAdisease associations with a high level of reliability.
Discussion
Accumulated experimental evidences have shown that lncRNAs play an important role in the human complex disease mechanism, and mutations or disorders of lncRNAs are associated with various complex diseases. More and more evidences show that it is crucial to propose an effective computational model to infer potential lncRNAdisease associations. In this article, we proposed a novel computational model of Laplacian normalization and birandom walks on heterogeneous networks for predicting lncRNAdisease associations. Our method shows better performance in LOOCV experiments by comparison with previous methods. In 50 unverified lncRNAdisease associations experiments, We compared our method with LRLSLDA, GrwLDA. The results indicated that our method has the higher ranking. Furthermore, the study of the cases of breast cancer, lung cancer, and liver cancer shows that our method improves the performance of predicting potential relationships.
Although our method can improve the prediction accuracy, it still has some limitations. For example, construction of the diseasedisease similarity matrix relies on the Gaussian interaction profile kernel similarity matrix for diseases from the known diseaselncRNA associations. In further work, we will improve our method in the following aspects: Firstly, LapBiRWRHLDA relies on the calculation of similarity matrix when constructing an lncRNA similarity network, and so the incompleteness of data may affect the final performance. Therefore, the integration of gene disease correlation data or the addition of more bioinformatics data may improve the performance of our method. These aspects have been considered in previous methods such as TPGLDA [9] and BRWLDA [15]. Secondly, the birandom algorithm performs random walk restarts on lncRNA similarity networks and disease similarity networks separately; how to better integrate random walks on two networks is an issue in our future research.
Conclusions
In this study, we proposed a method called LapBiRWRHLDA to predict the relationship between lncRNA and diseases. This model utilizes the Laplacian normalization of the lncRNA similarity matrix and the disease similarity matrix. Then constructs a heterogeneous network based on lncRNA similarity network, disease similarity network and available lncRNAdisease associations. Next, it applies birandom walks on the heterogeneous network to predict potential associations between lncRNAs and diseases. Our method can be used to better identify potential associations between lncRNAs and diseases.
The reason why our method has good results is mainly due to two factors. On the one hand, we exploit the similarity of lncRNAs by integrating Gaussian interaction profile kernel similarity of lncRNA and lncRNA expression similarity, and then apply Laplacian normalization. We also rely on lncRNA similarity matrices to construct an lncRNA similarity network. On the other hand, the birandom walk algorithm simulates random walk restarts on the lncRNA similarity network and disease similarity network; we then infer the relationship between lncRNAs and diseases by weighted averaging. We believe that the higher the score of potential lncRNAdisease relationship is, the higher the probability of association is.
Methods
Data sets
We downloaded three data sets of lncRNAdisease associations from the supplementary files of published articles [4, 8], which contains 293 experimentally confirmed lncRNAdisease relationships between 167 diseases and 118 lncRNAs from the LncRNADisease database on October 2012 [4], 454 known lncRNAdisease associations between 162 diseases and 187 lncRNAs from the LncRNADisease database on April 2016, and 594 lncRNAdisease associations between 79 diseases and 310 lncRNAs from the Lnc2Cancer database on July 2016 [8]. The adjacency matrix of lncRNAdisease associations is denoted as A, where the value A(i,j) of row i and column j is 1 if disease d(i) is related to lncRNA l(j), otherwise it is 0. Let L={l(1),l(2),â‹¯,l(nl)} denote the set of lncRNAs, and D={d(1),d(2),â‹¯,d(nd)} denote the set of diseases.
We also downloaded lncRNA expressions and the gene expression levels from the supplementary files of the published articles [4, 8], which contain 21626 expression profiles across 22 human tissues or cell types and 60245 gene expression levels in 16 tissues. Let set L_{1}, where L_{1} is composed of lncRNAs with lncRNA expression profiles (L_{1}âŠ†L). According to the previous approaches [4], if l(i), l(j)âˆˆL_{1}, we calculated the Spearman correlation coefficient of l(i) and l(j) as the lncRNA expression similarity. The lncRNA expression similarity matrix is represented by matrix SPC, where SPC(l(i),l(j)) is the expression similarity between l(i) and l(j) if they belongs to L_{1}, otherwise 0.
Laplacian normalization
Suppose that M=M(i,j),i,j=1,2,â‹¯,N, is a symmetric matrix, D is a diagonal matrix of which D(i,i) is the sum of row i of M and D(i,j)=0 for iâ‰ j. M is normalized by \(\hat {M} =D^{1/2}MD^{1/2}\), which also yields a symmetric matrix. The elements of \( \hat {M}\) are defined by
This process is called Laplacian normalization of M. It is often used to normalize a weighted matrix of a network [4, 20, 21].
Construction of the lncRNAlncRNA similarity matrix
Based on the assumption that similar diseases tend to show a similar interaction or noninteraction with the lncRNAs, the Gaussian interaction profile kernel similarity of lncRNAs can be calculated from known lncRNAdisease associations [4]. The lncRNA interaction profile IP(l(i)) is a binary vector which is 1 if lncRNA l(i) is related to the disease, 0 otherwise, defined as the ith column of the adjacency matrix A of the known lncRNAdisease association network constructed above. Then we can calculate the Gaussian interaction profile kernel similarity of lncRNA l(i) and lncRNA l(j) from their interaction profiles as
where the parameter Î³_{l} controls the kernel bandwith, which is calculated based on the new kernel bandwidth parameter \(\gamma _{l}^{\prime } \) as follows:
where nl denotes the number of lncRNAs. For simplicity we set \(\gamma _{l}^{\prime }=1\) as in the previous works [4, 22].
Following previous approaches [4], we construct the similarity of lncRNAs by combining the lncRNA expression similarity and Gaussian interaction profile kernel similarity. We denote by SL the lncRNA similarity matrix, where the element SL(i,j) defines the similarity between lncRNA l(i) and lncRNA l(j) as
where SPC(l(i),l(j)) represents the expression profile similarity of lncRNA l(i) and lncRNA l(j), and its value is the Spearman correlation coefficient of lncRNA l(i) and lncRNA l(j), so the matrix SPC is a symmetric matrix. KL(l(i),l(j)) represents the Gaussian interaction profile kernel similarity of lncRNA l(i) and lncRNA l(j), so the matrix KL is also a symmetric matrix. Therefore the lncRNA similarity matrix SL is a symmetric matrix. In Eq. (4), ew is the weight coefficient of lncRNA expression similarity; for simplicity we set ew=1/2.
Next, using Laplacian normalization, the element SL(i,j) is calculated through two steps:
Construction of the diseasedisease similarity matrix
Similar to lncRNAs, the Gaussian interaction profile kernel similarity of diseases can be constructed as
Here IP(d(i)) is defined as the ith row of the adjacency matrix A of the known lncRNAdisease association. It is a binary vector representing the relationship between disease d(i) and each gene. The Gaussian interaction profile kernel similarity matrix KD is a symmetric matrix. The parameter Î³_{d} is calculated as
where nd denotes the number of diseases; for simplicity we set \(\gamma _{d}^{\prime }=1\) as in the previous works [4, 22].
From relevant research [4, 19], to improve the predictive accuracy of disease similarity, we apply the logistic function transformation to represent the similarity of diseases. The disease similarity is redefined as
where c and d are two parameters, for which we adopt the same parameter selection as in the previous studies [4, 19], i.e. c=âˆ’15,d= log(9999). The disease similarity matrix SD is a symmetric matrix. Next, using Laplacian normalization, the element SD(i,j) is calculated through two steps:
Construction of the heterogeneous network
We first use the two matrices LD, LL to construct two networks, namely a disease similarity network, and an lncRNA similarity network. In the lncRNA similarity network, the edge between l(i) and l(j) is weighted by the similarity value of these two lncRNAs.
Likewise, in the disease similarity network, the edge between d(i) and d(j) is weighted by the similarity value of these two diseases.
Besides, the lncRNAdisease association network can be modeled as a bipartite graph. In this graph, the heterogeneous nodes correspond to either lncRNA or disease, and edges denote the presence or absence of the associations between them. If there is a known association between disease d(i) and lncRNA l(j), the weight of the edge is 1; otherwise it is 0. We divide the nodes of the heterogeneous network into two types. Those nodes connecting the lncRNA similarity network with the disease similarity network are called bridging nodes, and the other nodes are named internal nodes [21].
The heterogeneous network can be constructed by connecting the lncRNA similarity network and the disease similarity network via the known lncRNAdisease associations. A simple example of a heterogeneous network is illustrated in Fig. 4.
LapBiRWHLDA
In this study, we develop a novel computational method called LapBiRWHLDA to predict human lncRNAdisease associations. Figure 5 shows the flowchart of LapBiRWHLDA. Firstly, lncRNA similarity and disease similarity can be calculated based on the known lncRNAdisease associations taken from the LncRNADisease database. Secondly, the global heterogeneous network is built by combining the lncRNA similarity network, the disease similarity network and the lncRNAdisease association network. Finally, the birandom walk algorithm is performed on the heterogeneous network to obtain the association probability scores between lncRNAs and diseases.
Suppose a random walker can jump from d(1) to l(1) and then to l(2). We can take d(1) as the starting node for the random walk. To simulate this process, we apply a random walk on the lncRNA similarity network. The iterative process can be described as
Similarly, we can also apply a random walk on the disease similarity network as follows:
where Î± is a parameter to control the restart probability for the random walker, \({RT}_{L}^{t}\) is the predicted association between lncRNA l and disease d in the tth iteration, \({RL}_{D}^{t}\) is the predicted relevance between disease d and lncRNA l in the tth iteration, with
After the bi random walks in the disease similarity network and in the lncRNA similarity network in the tth step, LapBiRWHLDA further combines \( {RT}_{L}^{t}\) and \({RT}_{D}^{t}\) into RT^{t} as follows:
After several steps, when the change between RT^{t+1} and RT^{t} is less than 10^{âˆ’10}, we obtain the steady prediction score matrix RT, where RT(i,j) is the probability of potential association disease d(i) and lncRNA l(j).
Abbreviations
 AUC:

Areas under ROC curve
 LapBiRWRHLDA:

Laplacian normalization and birandom walks on heterogeneous networks for predicting lncRNAdisease associations
 LOOCV:

Leaveoneout crossvalidation
 ROC:

Receiveroperating characteristics
References
Kapranov P, Cheng J, Dike S, Nix DA, Duttagupta R, Willingham AT, Stadler PF, Hertel J, Hackermller J, Hofacker IL, Bell I, Cheung E, Drenkow J, Dumais E, Patel S, Helt G, Ganesh M, Ghosh S, Piccolboni A, Sementchenko V, Tammana H, Gingeras TR. RNA maps reveal new RNA classes and a possible function for pervasive transcription. Science. 2007; 316(5830):1484â€“8.
Mercer TR, Dinger ME, Mattick JS. Long noncoding RNAs: insights into functions. Nat Rev Genet. 2009; 10(3):155â€“9.
Wapinski O, Chang HY. Long noncoding RNAs and human disease. Trends Cell Biol. 2011; 21(6):354â€“61.
Chen X, Yan GY. Novel human lncRNAdisease association inference based on lncRNA expression profiles. Bioinformatics. 2013; 29(20):2617â€“24.
Chen X, Yan CC, Luo C, et al. Constructing lncRNA functional similarity network based on lncRNAdisease associations and disease semantic similarity. Sci Rep. 2015; 5:11338.
Chen X, Huang YA, Wang XS, et al. FMLNCSIM: fuzzy measurebased lncRNA functional similarity calculation model. Oncotarget. 2016; 7(29):45948â€“58.
Chen X. KATZLDA: KATZ measure for the lncRNAdisease association prediction. Sci Rep. 2014; 5:16840.
Peng H, Lan C, Liu Y, et al. Chromosome preference of disease genes and vectorization for the prediction of noncoding disease genes. Oncotarget. 2017; 8(45):78901â€“16.
Ding L, Wang M, Sun D, et al. TPGLDA: Novel prediction of associations between lncRNAs and disease via lncRNAdiseasegene tripartite graph. Sci Rep. 2018; 8(1):1065.
Chen X. Predicting lncRNAdisease associations and constructing lncRNA functional similarity network based on the information of miRNA[J]. Sci Rep. 2015; 5:13186.
Zhou M, Wang X, Li J, et al. Prioritizing candidate diseaserelated long noncoding RNAs by walking on the heterogeneous lncRNA and disease network. Mol BioSyst. 2014; 11(3):760.
Chen X, You ZH, Yan GY, et al. IRWRLDA: improved random walk with restart for lncRNAdisease association prediction. Oncotarget. 2016; 7(36):57919â€“31.
Sun J, Shi H, Wang Z, et al. Inferring novel lncRNAdisease associations based on a random walk model of a lncRNA functional similarity network. Mol BioSyst. 2014; 10(8):2074â€“81.
Gu C, Li XY, Cai LJ, et al. Global network random walk for predicting potential human lncRNAdisease associations. Sci Rep. 2017; 7(1):12442.
Yu G, Fu G, Lu C, et al. BRWLDA: birandom walks for predicting lncRNAdisease associations. Oncotarget. 2017; 8(36):60429â€“46.
Donahue HJ, Genetos DC. Genomic approaches in breast cancer research. Brief Funct Genom. 2013; 12(5):391â€“6.
Karagoz K, Sinha R, Arga KY. Triple Negative Breast Cancer: A MultiOmics Network Discovery Strategy for Candidate Targets and Driving Pathways. Omicsa J Integr Biol. 2015;19(2).
Bosch FX, Ribes J, Borrs J. Epidemiology of Primary Liver Cancer. Sem Liver Dis. 1999; 19(03):271â€“85.
Center MM, Jemal A. International trends in liver cancer incidence rates. Cancer Epidemiol Biomarkers Prev. 2011; 20(11):2362â€“8.
Vanunu O, Magger O, Ruppin E, et al. Associating genes and protein complexes with disease via network propagation. PLoS Comput Biol. 2010; 6(1):e1000641.
Zhao ZQ, Han GS, Yu ZG, Li JY. Laplacian normalization and random walk on heterogeneous networks for diseasegene prioritization. Comput Biol Chem. 2015; 57(C):21â€“28.
Van LT, Nabuurs SB, Marchiori E. Gaussian interaction profile kernels for predicting drugtarget interaction. Bioinformatics. 2011; 27(21):3036.
Acknowledgements
The authors thank the anonymous referees, especially Prof. Jinyan Li at the University of Technology Sydney, for suggestions that helped improve the paper substantially.
Funding
Publication of this article was sponsored by the Natural Science Foundation of China (Grant No.11401503), Natural Science Foundation of Hunan Province of China (Grant No. 2016JJ3116), Outstanding Youth Foundation of Hunan Educational Committee (Grant No.16B256).
Availability of data and materials
The data sets used in this study are available from the corresponding author on reasonable request.
About this supplement
This article has been published as part of BMC Systems Biology Volume 12 Supplement 9, 2018: Proceedings of the 29th International Conference on Genome Informatics (GIW 2018): systems biology. The full contents of the supplement are available online at https://bmcsystbiol.biomedcentral.com/articles/supplements/volume12supplement9.
Author information
Authors and Affiliations
Contributions
YPW and GSH contributed to the conception and design of the study and developed the method. YPW implemented the algorithms and analyzed the data and results. GSH gave the ideas and supervised the project. YPW wrote the manuscript, GSH and Vo V Anh reviewed the final manuscript. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Publisherâ€™s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
About this article
Cite this article
Wen, Y., Han, G. & Anh, V.V. Laplacian normalization and birandom walks on heterogeneous networks for predicting lncRNAdisease associations. BMC Syst Biol 12 (Suppl 9), 122 (2018). https://doi.org/10.1186/s1291801806600
Published:
DOI: https://doi.org/10.1186/s1291801806600
Keywords
 Birandom walk
 Disease similarity network
 Laplacian normalization
 LncRNA similarity network