MicroRNA Detection CS374 Spring 2008 Lecture 12, 5/8/08Lecturer: Khan Shing Scribe: Jason Turner-MaierMicroRNA DetectionBased on the following papers:1. Stark A., Kheradpour, P., Parts, L., Brennecke, J., Hodges, E., Hannon, G., and Kellis, M. “Sys-tematic discovery and characterization of fly microRNAs using 12 Drosophila genomes”, GenomeResearch, 17(12):1865-1879, 2007.2. Kertesz, M., Iovino, N., Unnerstall, U., Gaul, U., and Segal, E. “The role of site accessibility inmicroRNA target recognition”, Nature Genetics, 39(10):1278-1284, 2007.Additional references:1. Zamore, P.D. and Haley, B. "Ribo-gnome: The big world of small RNAs", Science, 309:1519–1524, 2005.1. IntroductionMicroRNAs (miRNA) regulate gene expression. Genes, when expressed, cause the for-mation of all the different proteins in living organisms. Since proteins accomplish domuch of the work associated with life (such as motility and reaction catalysis), any in-formation about how, why, and when they are formed is vital to understanding any num-ber of different areas of biology. It is therefore important to be able to identify what se-quences in the genome form miRNAs to aid further study.2. Background2.1 MicroRNAProteins are a biomolecule vital to life. They perform any number of vital tasks, such assupporting cellular structure or mediating transport of molecules across a membrane. Allproteins in an organism are encoded by that organisms DNA via the genetic code. TheDNA is transcribed into messenger RNA (abbreviated mRNA), which is single stranded(unlike DNA). The mRNA is then used to form the corresponding protein by a processcalled translation.This process is highly regulated. A cell doesn't want to express all of its genes all of thetime, so it is important to be able to prevent some portion of them from becoming protein.For example, if a cell has a protein for breaking down sucrose, it may not want to producethis protein when there is no sucrose. If a cell has a protein for breaking down the nuclearmembrane, it likely doesn't want to produce this protein except when it is undergoingmitosis. There are several steps for turning a gene into a protein as outlined above, andregulation can occur at any of them. Transcriptional regulation affects whether the gene istranscribed into an mRNA strand. Post-transcriptional regulation affects whether anmRNA is translated into a protein. Post-translational regulation affects whether a trans-lated protein is active and whether it remains in the cell or is degraded.MiRNAs are a type of factor which affect post-transcriptional regulation. They are shortRNA sequences transcribed from the genome. It is estimated that 1-5% of animal genesMicroRNA Detection CS374 Spring 2008 Lecture 12, 5/8/08Lecturer: Khan Shing Scribe: Jason Turner-Maiermay code for miRNAs. RNA can have secondary and tertiary structure just like proteinscan, and miRNAs form a characteristic hairpin structure that is crucial for their furtherprocessing (Fig. 1). Only one half of the stem section of the miRNA is actually used; therest is spliced out post-transcriptionally (Fig. 2). The section which is used is highlightedin red below. MiRNAs are a relatively new discovery, but estimates suggest that theymay participate in regulation of > 30% of human genes.Fig. 1. MicroRNA structure (Source: Paper 1)Fig. 2. MicroRNA processing (Source: Addl. Reference 1)Before it can participate in regulation, the miRNA needs to have some processing per-formed. After the it is exported to the cytoplasm, the upper loop is spliced off. The re-sulting double-stranded RNA is separated, and one strand is then incorporated into aprotein complex called RNA induced silencing complex, or RISC. RISC target mRNAfor degradation, both by guiding cleavage of the mRNA and by promoting transport to aMicroRNA Detection CS374 Spring 2008 Lecture 12, 5/8/08Lecturer: Khan Shing Scribe: Jason Turner-Maierregion where the mRNA won't have an effect. However, it is the miRNA that is insertedinto RISC which gives it its specificity. The miRNA recognizes a complementary se-quence on the target mRNA, allowing RISC to only affect the desired mRNA molecules.2.2 Decision TreesA decision tree is a method of encoding a way of choosing between options. That is, adecision tree is designed to output a choice given some amount of data. For example, adecision tree could be designed to decide whether a given animal was a dog, a cat, a wolf,a fish, or a bird given information such as whether the animal has fur or whether it flies.As the name implies, a decision tree has as its structure a tree. Each leaf node is a choice,and each non-leaf node encodes a means of deciding between child nodes of that nodegiven some data. For example, a leaf node might be "cat", and a non-leaf node might be:"If it has fur, choose the right child. Otherwise, choose the left." When using a decisiontree, you start with the root node, and traverse downwards taking the path as indicated bythe non-leaf nodes. When you reach a leaf node, you have your answer.For example, consider the tree below (Fig. 3).Fig. 3: Example Decision Tree (Source: http://www.gmupolicy.net/its/incidentduration/image351.gif)If this were a tree for solving the above problem, it might have the following assign-ments:X1: If it has fur, choose the right child. Otherwise, choose the left.X2: If it flies, choose the right child. Otherwise, choose the left.X3: If it is commonly domesticated, choose the right child. Otherwise, choose the left.X4: If it always lands on its feet, choose the right child. Otherwise choose the left.Terminal Node 1: FishTerminal Node 2: BirdTerminal Node 3: WolfTerminal Node 4: DogTerminal Node 5: CatMicroRNA Detection CS374 Spring 2008 Lecture 12, 5/8/08Lecturer: Khan Shing Scribe: Jason Turner-MaierNote that the assumption here is that we have an animal, and we know things about thatanimal, but we don't know what type it is. Suppose we have a furred non-domesticatedanimal which does not fly and does not land on its feet. We would start from the rootnode, and move right, since it has fur. Then we would move left, since it is not domesti-cated. We would then be at a leaf node, and would be able to conclude that the animal is awolf.Decision trees are commonly built using machine learning techniques, rather than byhand. To do this, we need a training set of labeled examples. The goal is to subdivide theexamples using data about them,
View Full Document
Unlocking...