DWNN-RLS: regularized least squares method for predicting circRNA-disease associations

Abstract Background Microbes are closely related to human health and diseases. Identification of disease-related microbes is of great significance for revealing the pathological mechanism of human diseases and understanding the interaction mechanisms between microbes and humans, which is also useful for the prevention, diagnosis and treatment of human diseases. Considering the known disease-related microbes are still insufficient, it is necessary to develop effective computational methods and reduce the time and cost of biological experiments. Methods In this work, we developed a novel computational method called MDAKRLS to discover potential microbe-disease associations (MDAs) based on the Kronecker regularized least squares. Specifically, we introduced the Hamming interaction profile similarity to measure the similarities of microbes and diseases besides Gaussian interaction profile kernel similarity. In addition, we introduced the Kronecker product to construct two kinds of Kronecker similarities between microbe-disease pairs. Then, we designed the Kronecker regularized least squares with different Kronecker similarities to obtain prediction scores, respectively, and calculated the final prediction scores by integrating the contributions of different similarities. Results The AUCs value of global leave-one-out cross-validation and 5-fold cross-validation achieved by MDAKRLS were 0.9327 and 0.9023 ± 0.0015, which were significantly higher than five state-of-the-art methods used for comparison. Comparison results demonstrate that MDAKRLS has faster computing speed under two kinds of frameworks. In addition, case studies of inflammatory bowel disease (IBD) and asthma further showed 19 (IBD), 19 (asthma) of the top 20 prediction disease-related microbes could be verified by previously published biological or medical literature. Conclusions All the evaluation results adequately demonstrated that MDAKRLS has an effective and reliable prediction performance. It may be a useful tool to seek disease-related new microbes and help biomedical researchers to carry out follow-up studies.

Download Full-text

Accelerating migration deconvolution using a nonmonotone gradient method

Geophysics ◽

10.1190/1.3457923 ◽

2010 ◽

Vol 75 (4) ◽

pp. S131-S137 ◽

Cited By ~ 19

Author(s):

Yanfei Wang ◽

Changchun Yang

Keyword(s):

Least Squares ◽

Gradient Descent ◽

Least Squares Method ◽

Adjoint Operator ◽

Three Dimensional ◽

Descent Method ◽

Gradient Descent Method ◽

Seismic Migration ◽

Regularized Least Squares ◽

Practical Applications

New solution methods were considered for migration deconvolution in seismic imaging problems. It is well known that direct migration methods, using the adjoint operator [Formula: see text], yield a lower-resolution or blurred image, and that the linearized inversion of seismic data for the reflectivity model usually requires solving a (regularized) least-squares migration problem. We observed that the (regularized) least-squares method is computationally expensive, which becomes a severe obstacle for practical applications. Iterative gradient-descent methods were studied and an efficient method for migration deconvolution was developed. The problem was formulated by incorporating regularizing constraints, and then a nonmonotone gradient-descent method was applied to accelerate the convergence. To test the potential of the application of the developed method, synthetic two-dimensional and three-dimensional seismic-migration-deconvolution simulations were performed. Numerical performance indicates that this method is promising for practical seismic migration imaging.

Download Full-text