Molecular biomarkers are certain molecules or set of molecules that can be of help for diagnosis or prognosis of diseases or disorders. In the past decades, thanks to the advances in high-throughput technologies, a huge amount of molecular ‘omics’ data, e.g. transcriptomics and proteomics, have been accumulated. The availability of these omics data makes it possible to screen biomarkers for diseases or disorders. Accordingly, a number of computational approaches have been developed to identify biomarkers by exploring the omics data. In this review, we present a comprehensive survey on the recent progress of identification of molecular biomarkers with machine learning approaches. Specifically, we categorize the machine learning approaches into supervised, un-supervised and recommendation approaches, where the biomarkers including single genes, gene sets and small gene networks. In addition, we further discuss potential problems underlying bio-medical data that may pose challenges for machine learning, and provide possible directions for future biomarker identification.

Although a lot of methods have been proposed to identify driver genes, how to separate the driver mutations from the passenger mutations is still a challenging problem in cancer genomics. The detection of driver genes with rare mutation and low accuracy is unsolved better. In this study, we present an integrated network-based approach to locate potential driver genes in a cohort of patients. The approach is composed of two steps including a network diffusion step and an aggregated ranking step, which fuses the correlation between the gene mutations and gene expression, the relationship between the mutated genes and the heterogeneous characteristic of the patient mutation. We analyze three cancer datasets including Glioblastoma multiforme, Ovarian cancer and Breast cancer. Our method has not only identified the known driver genes with high-frequency mutations, but also discovered the potential driver genes with a rare mutation. At the same time, validation by literature search and functional enrichment analysis reveal that the predicted genes are obviously related to these three kinds of cancers.

The interactions between the microbiota and the human host can affect the physiological functions of organs (such as the brain, liver, gut, etc.). Accumulating investigations indicate that the imbalance of microbial community is closely related to the occurrence and development of diseases. Thus, the identification of potential links between microbes and diseases can provide insight into the pathogenesis of diseases. In this study, we propose a deep learning framework (MDAGCAN) based on graph convolutional attention network to identify potential microbe-disease associations. In MDAGCAN, we first construct a heterogeneous network consisting of the known microbe-disease associations and multi-similarity fusion networks of microbes and diseases. Then, the node embeddings considering the neighbor information of the heterogeneous network are learned by applying graph convolutional layers and graph attention layers. Finally, a bilinear decoder using node embedding representations reconstructs the unknown microbe-disease association. Experiments show that our method achieves reliable performance with average AUCs of 0.9778 and 0.9454 ± 0.0038 in the frameworks of Leave-one-out cross validation (LOOCV) and 5-fold cross validation (5-fold CV), respectively. Furthermore, we apply MDAGCAN to predict latent microbes for two high-risk human diseases, i.e., liver cirrhosis and epilepsy, and results illustrate that 16 and 17 out of the top 20 predicted microbes are verified by published literatures, respectively. In conclusion, our method displays effective and reliable prediction performance and can be expected to predict unknown microbe-disease associations facilitating disease diagnosis and prevention.

The identification of deregulated modules (such as induced by oncogenes) is a crucial step for exploring the pathogenic process of complex diseases. Most of the existing methods focus on deregulation of genes rather than the links of the path among them. In this study, we emphasize on the detection of deregulated links, and develop a novel and effective regulatory path-based approach in finding deregulated modules. Observing that a regulatory pathway between two genes might involve in multiple rather than a single path, we identify condition-specific core regulatory path (CCRP) to detect the significant deregulation of regulatory links. Using time-series gene expression, we define the regulatory strength within each gene pair based on statistical dependence analysis. The CCRPs in regulatory networks can then be identified using the shortest path algorithm. Finally, we derive the deregulated modules by integrating the differential edges (as deregulated links) of the CCRPs between the case and the control group. To demonstrate the effectiveness of our approach, we apply the method to expression data associated with different states of Human Epidermal Growth Factor Receptor 2 (HER2). The experimental results show that the genes as well as the links in the deregulated modules are significantly enriched in multiple KEGG pathways and GO biological processes, most of which can be validated to suffer from impact of this oncogene based on previous studies. Additionally, we find the regulatory mechanism associated with the crucial gene SNAI1 significantly deregulated resulting from the activation of HER2. Hence, our method provides not only a strategy for detecting the deregulated links in regulatory networks, but also a way to identify concerning deregulated modules, thus contributing to the target selection of edgetic drugs.

The gut microbiota plays a vital role in human health, and significant effort has been made to predict human phenotypes, especially diseases, with the microbiota as a promising indicator or predictor with machine learning (ML) methods. However, the accuracy is impacted by a lot of factors when predicting host phenotypes with the metagenomic data, e.g. small sample size, class imbalance, high-dimensional features, etc. To address these challenges, we propose MicroHDF, an interpretable deep learning framework to predict host phenotypes, where a cascade layers of deep forest units is designed for handling sample class imbalance and high dimensional features. The experimental results show that the performance of MicroHDF is competitive with that of existing state-of-the-art methods on 13 publicly available datasets of six different diseases. In particular, it performs best with the area under the receiver operating characteristic curve of 0.9182 ± 0.0098 and 0.9469 ± 0.0076 for inflammatory bowel disease (IBD) and liver cirrhosis, respectively. Our MicroHDF also shows better performance and robustness in cross-study validation. Furthermore, MicroHDF is applied to two high-risk diseases, IBD and autism spectrum disorder, as case studies to identify potential biomarkers. In conclusion, our method provides an effective and reliable prediction of the host phenotype and discovers informative features with biological insights.