Chen et al. [58] proposed a semi-supervised learning
framework to predict lncRNA-disease associations. It integrated
lncRNA expression similarity and Gaussian interaction profile kernel
similarity for lncRNA. The Laplacian Regularized Least Squares
method was employed for lncRNA-disease inference. In consider-
ation of the important of lncRNA similarity, Chen et al. [59] pro-
posed two novel lncRNA functional similarity calculation models
based on disease information. The Laplacian Regularized Least
Squares was used for lncRNA-disease association prediction. As a
result, new predictive models improved the performance of
LRLSLDA in the leave-one-out cross validation of various known
lncRNA-disease association datasets. Similar works have been devel-
oped by Huang et al. [60] and Chen et al. [61]. Huang et al. [60]
presented an improved lncRNA functional similarity calculation
method (ILNCSIM) based on information content method. The
main improvement was that it combines hierarchical structure of
disease directed acyclic graphs and information content for disease
similarity calculation. The Laplacian Regularized Least Squares was
employed for lncRNA-disease association identification. The results
showed that it can enhance the accuracy of prediction than previous
method. Chen et al. [61] developed a fuzzy measure-based lncRNA
functional similarity calculation model (FMLNCSIM) to identify
interactions between lncRNAs and diseases. In this work, Sugeno
λ-measure was used to calculate disease semantic similarity based on
Medical Subject Headings (MeSH). Then, the Laplacian Regular-
ized Least Squares was employed to evaluate the effectiveness of
FMLNCSIM. The results demonstrated that FMLNCSIM could be
used for searching functionally similar lncRNAs.
In addition, Biswas et al. [62] proposed a computational frame-
work for lncRNA-disease interaction prediction based on matrix
factorization. This work integrated lncRNA-disease association,
experimentally validated gene-disease association, gene-gene inter-
action data, and expression profiles of both lncRNAs and genes.
The integrative non-negative matrix factorization method was used
to infer lncRNA-disease interactions, and bi-cluster was employed
to identify the module of lncRNA. Experimental results showed the
superiority of our proposed method over two state-of-the-art clus-
tering algorithms—k-means and hierarchical clustering.
The challenge of machine learning method is how to select
useful biological features to train classifier. Therefore, integrating
multiple data resource is an effective method to improve perfor-
mance. However, the redundant or irrelevant biological informa-
tion may be futile even degrade the performance. How to select
useful features from various biological resources would be another
research focus [63, 64]. In addition, the different classification
algorithms may be suit for different data resources. Hence, utilizing
multiple learning algorithms can obtain better predictive
performance.Identifying Interactions Between Long Noncoding RNAs and Diseases... 215