A data mining approach to discover unusual folding regions in genome sequences

Segmentation of DNA using simple recurrent neural network

We report the discovery of strong correlations between protein coding regions and the prediction errors when using the simple recurrent network to segment genome sequences. We are going to use SARS genome to demonstrate how we conduct training and derive corresponding results. The distribution of prediction error indicates how the underlying hidden regularity of the genome sequences and the results are consistent with the finding of biologists: predicated protein coding features of SARS genome. This implies that the simple recurrent network is capable of providing new features for further biological studies when applied on genome studies. The HA gene of influenza A subtype H1N1 is also analyzed in a similar way.

Two-tiered approach identifies a network of cancer and liver disease-related genes regulated by miR-122

MicroRNAs function as important regulators of gene expression and are commonly linked to development, differentiation, and diseases such as cancer. To better understand their roles in various biological processes, identification of genes targeted by microRNAs is necessary. Although prediction tools have significantly helped with this task, experimental approaches are ultimately required for extensive target search and validation. We employed two independent yet complementary high throughput approaches to map a large set of mRNAs regulated by miR-122, a liver-specific microRNA implicated in regulation of fatty acid and cholesterol metabolism, hepatitis C infection, and hepatocellular carcinoma. The combination of luciferase reporter-based screening and shotgun proteomics resulted in the identification of 260 proteins significantly down-regulated in response to miR-122 in at least one method, 113 of which contain predicted miR-122 target sites. These proteins are enriched for functions associated with the cell cycle, differentiation, proliferation, and apoptosis. Among these miR-122-sensitive proteins, we identified a large group with strong connections to liver metabolism, diseases, and hepatocellular carcinoma. Additional analyses, including examination of consensus binding motifs for both miR-122 and target sequences, provide further insight into miR-122 function.

Data mining of functional RNA structures in genomic sequences

The normal functions of genomes depend on the precise expression of messenger RNAs and noncoding RNAs (ncRNAs) such as transfer RNAs and microRNAs in eukaryotes. These ncRNAs and functional RNA structures (FRSs) act as regulators or response elements for cellular factors and participate in transcription, posttranscriptional processing, and translation. Knowledge discovery of these FRSs in huge DNA/RNA sequence databases is a very important step to reach our goal of going from genomic sequence data to biological knowledge for understanding RNA-based regulation. Analyses of a large number of FRSs have indicated that the FRS can be well characterized by some quantitative measures such as significance and well-ordered scores of the local segment. Various data mining tools have been developed and successfully applied to FRS discovery in genomic sequence databases. Here, we summarize our efforts in the computational discovery of structured features of ncRNAs and FRSs within complex genomes by EDscan and SigED.