Background: High-throughput bio-OMIC technologies are producing high-dimension data from bio-samples at an ever increasing rate, whereas the training sample number in a traditional experiment remains small due to various difficulties. This “large p, small n” paradigm in the area of biomedical “big data” may be at least partly solved by feature selection algorithms, which select only features significantly associated with phenotypes. Feature selection is an NP-hard problem. Due to the exponentially increased time requirement for finding the globally optimal solution, all the existing feature selection algorithms employ heuristic rules to find locally optimal solutions, and their solutions achieve different performances on different datasets. Results: This work describes a feature selection algorithm based on a recently published correlation measurement, Maximal Information Coefficient (MIC). The proposed algorithm, McTwo, aims to select features associated with phenotypes, independently of each other, and achieving high classification performance of the nearest neighbor algorithm. Based on the comparative study of 17 datasets, McTwo performs about as well as or better than existing algorithms, with significantly reduced numbers of selected features. The features selected by McTwo also appear to have particular biomedical relevance to the phenotypes from the literature. Conclusion: McTwo selects a feature subset with very good classification performance, as well as a small feature number. So McTwo may represent a complementary feature selection algorithm for the high-dimensional biomedical datasets.
Clustered regularly interspaced short palindromic repeats (CRISPRs) are important genetic elements in many bacterial and archaeal genomes, and play a key role in prokaryote immune systems’ fight against invasive foreign elements. The CRISPR system has also been engineered to facilitate target gene editing in eukaryotic genomes. Using the common features of mis-annotated CRISPRs in prokaryotic genomes, this study proposed an accurate de novo CRISPR annotation program CRISPRdigger, which can take a partially assembled genome as its input. A comprehensive comparison with the three existing programs demonstrated that CRISPRdigger can recover more Direct Repeats (DRs) for CRISPRs and achieve a higher accuracy for a query genome. The program was implemented by Perl and all the parameters had default values, so that a user could annotate CRISPRs in a query genome by supplying only a genome sequence in the FASTA format. All the supplementary data are available at http://www.healthinformaticslab.org/supp/.
Chronic diseases have seriously affected human activities, especially in many developing countries and underdeveloped countries. The long duration of chronic diseases and the high cost of medical care have placed a huge economic burden on society and families. Meanwhile, chronic patients tend to have a variety of complications over time. So, it is difficult for doctors to find effective diagnosis and appropriate treatment. Machine learning techniques can integrate their heterogeneous data of various body indicators. Meanwhile, for chronic patients, multi-label learning methods can be used to help doctors identify the types of the chronic diseases. This paper proposes a novel multi-label neural network method (ML-NN) to predict the chronic diseases combining neural network and multi-label learning technology based on cross entropy lost function and backward propagation algorithm. Compared with 14 traditional multi-label learning methods on 10 chronic diseases and 19733 patients, the proposed method achieved a consistently best in 5 performance measurements. The results demonstrate the proposed method can effectively predict chronic diseases and assist doctors to diagnose and treat patients.