Bioinformatics: Introduction and Methods (Week 6)

# Week6 Variant Database Statement before class: All the contents here marked are belong to the course [Bioinformatics: Introduction and Methods](https://www.coursera.org/learn/bioinformatics-pku/home/welcome). If you like this high-quality course contents, please turn to the official course for more details. Here, i recored the knowledge learnt for convenience. If any copyright problem exist, please tell me, i’ll delete all immediately, Thanks! ## Overview of rhe Problem ### 1. Genetic Variations ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn3u9w2ayj30qy0gu134.jpg) - chromosomal aneuploidy; - structual variations (SVs); - Indel; - SNV (Single Nucleotide Variation: about 3 minllion SNVs in one person's genome, equivalent ~1/1000 frequency); - SNVs within coding regions (stop codon gain-> nonsense, non-synonymous -> missense, synonymous -> same sense/silent, affect splicing, stop codon loss) ### 2. Sum - `Nonesense SNVS are usually considered deleterious` - even though is is not always the case... - `synonymous, intronic, and intergenic variations are often ignored.` - GWAS show 88% of trait-associated variants of weak effect are non-coding - remain unclear-studied and new methods are needed - `Most studied so far had focused on missense SNVS` - `Known deleterious mutations are enriched in missense mutations` - ~50 of all known mutations of Medelian discorders are missense mutations - `Although missense SNVs changen the protein seq, many do not cause phrnotypic changes` - On average, a healty indicidual has | Class | Number | | :-----------------------: | :-------: | | SNP | 3,019,909 | | Indel | 361,669 | | Deletions | 15,893 | | Deplications | 407 | | Mobile element insertions | 4,775 | - While in protein-coding regions | Calss | Number | | :--------------------------------: | :------: | | Genes disturped by large deletions | 147 | | Stop-introducing SNPs | 1,057 | | Stop losses | 77 | | Small frameshift indels | 954 | | Small in-frome indels | 714 | | Non-synonymous SNPs | 68,300 | | Synonymous SNPs | 60,157 | ## Variant Database ### 1. The development of common variarnt DBs ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn4tnz5bmj30x009kjwd.jpg) ## `Functional predicton of genetic variants` ### `1. Concervation-based methos: SIFT` SIFT: Sort Intolerant From Tolerant subtitutions (http://sift/jcvi.org/). `① Mutations in important positions (eg active sites) tend to be conserve, tend to be deleterious; ② Mutations in positons that have high degree of diversity across specie, tend to be neutral. ` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn5gcffj7j30wg0hcgu8.jpg) ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn5jg55a8j30x20dwkfg.jpg) ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn5o4x5krj30xi0hgqfm.jpg) `Accuacy of SIFT: False negtive rate-> 31%, False positive rate -> 20R%, Coverage -> 60%` ### `2. The definication of accuacy` `This figure below explain accuacy well.` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn67c0xufj30xs0g6naj.jpg) ### `3. A rule-based method: PolyPhen` PolyPhen: Polymorphism Phenotyping (http://genetics.bwh.harvard.edu/pph2/), which predicts impact of AA variants based on both `multi-sequence alignmemnt and protein 3D structure features ` - AA variants at `concerved positions` are more likely to cause ` functional changes` - AA variants that affect `active sites, interaction sites, solubility, or stablity of a protein` are likely to affect protein structure. - Changes in protein structure are likely to casue in protein function, which are likely to cause changes in phenotype. - Empirically derived rules to predict damaging or begin ### `4. Claasifier-based method: SAPRED` SAPRED: Single Amino Acid Polymorphisms disease-associated Predictor (www.spared.cbi.pku.edu.cn). ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghn6tpbilzj30o40aoq6d.jpg) - PDB - get 3D protein structures - Homology Modeling - predict structure - Biologically -Intuitive features - `Structure neighbor profile: vector` - `Nearby functional sites` - `Hyfrogen bond change` <table><tr><td bgcolor=#ebffeb> `Wherever there are challenges, there are opprtunities.` </td></tr></table> ## Supplementary ### Introduction to SVM `Supervised machine learning: trainning data -> common machine learning methods(eg. SVM, HMM...) -> features -> new data -> model -> prediction` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghp94hi2l7j30q80dmtd5.jpg) `Classifying data is a common task in machine learning.` SVM: Support Vector Machines (回顾分析+模式识别). Three typical characters. - A surprised learning model -> `classification and regression analysis.` - Select a small of number of critical boundary instances (`surpport vectors` ) -> build `linear discriminant functions` -> seprate them as widely as possible. - `Kenerl trick` -> perform `non-linear classification` with mapping inputs into `high-dimensional fearture spaces.` 1. **Decision Boundary** Three key points will get most suitable line to seprate two spaces. Thus, this `most suitable line called decision boundary.` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghp9wzr2ilj30c00cwta5.jpg) 2. **Support Vector** Above ` three key points` , which used to get the only decision boundary for obtaning classificated spaces `called support vector.` And be free from discrete points. ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpa01ogthj30d60a8abd.jpg) 3. **Mathmatics** <center> The hyperplane is $$ w^Tx + b = 0 $$ </center> So, the classification function is <center> $$ f(x) = w^Tx + b \\\\ \\\\ y = \begin{cases} -x,\quad x\leq 0 \\\\ x,\quad x>0 \end{cases} $$ </center> ① Solve with `Quadratic Programming` <center> $$ min\frac{1}{2}||w||^2 \quad\quad s.t.y_i(w_Tx_i+b)≥ 1, \quad\quad i = 1,2,...,n. $$ </center> ② Solve with `Lagrange multipilers ` <center> $$ L(w,b,\alpha) = \frac{1}{2}||w||^2-\sum_{i=1}^{n}\alpha_i[y_i(w^Tx_i+b)-1] \\\\ \frac{\delta L}{\delta b} = 0 \quad=> \quad w = \sum_{i=1}^{n}\alpha_iy_ix_i \\\\ \frac{\delta L}{\delta b} = 0 \quad => \quad \sum_{i=1}^{n}\alpha_iy_i =0 $$ </center> `Finally, the calssification function con be rewrite as` <center> $$ f(x) = \left(\sum_{i=1}^{n} \alpha_iy_ix_i\right)^Tx+b \\\\ = \sum_{i=1}^{n}\alpha_iy_i(x_i,x)+b $$ </center> 4. **Kenel function (核函数)** Non-linear cases need more flexible `hypothetical space.` We can use `φ` to map `x` to a `higher dimension space, in which all the points can be linear separable.` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpaxa0z4kj30ek09w41j.jpg) So the classification function can be extened as: <center> $$ f(x) = \sum_{i=1}^{n}\alpha_iy_i(\Phi(x_i),\Phi(x))+b \\\\ $$ </center> Will get the `kernel function:` <center> $$ K(x,z) = (\Phi(x),\Phi(z)) $$ </center> ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpb2ciazkj30t40ewaez.jpg) <table><tr><td bgcolor=#ebffeb> `Common used kernel functions:`</td></tr></table> ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpb4axcsfj30pe0f0aec.jpg) 5. Apply SVM model in R ```R > data <- read.table("var.tsv", header =T) > head(data) > tail(data) -------------------------------->>>> > x <- subset(data, select = -class) > y <- factor(data$class) > library(e1071) -------------------------------->>>> > model <- svm(x, y, cost=2, gamma=0.5, probability=T, scale=F, cross=10) > summary(model) ### check the final "Total accuracy" in output -------------------------------->>>> > model$SV > pred <- predict(model, x, probability=T) > prob <- attr(pred, "probabilities")[,2] > head(prob) > prob -------------------------------->>>> > library(pROC) > roc <- roc(y, prob, smooth=T) > plot(roc, col="blue", print.thres=T, print.auc=T, grid=c(0.2,0.2)) -------------------------------->>>> > library(rgl) > x <- subset(x, select=c(a1,a2,a3)) > model <- svm(x, y, kernel="linear", cost=2, scale=F, cross=10) > palette(c("green", "blue")) > plot3d(x, col=as.integer(y), type="s", radius=0.05) > plot3d(model$SV, col=as.integer(y[model$index]), type="s", radius=0.15, add=T) -------------------------------->>>> > library(misc3d) > every.list <- lapply(x, function(x){seq(min(x),max(x),len=100)}) > every <- expand.grid(every.list) > pred <- predict(model, every, decision.values=T) > every.dv <- attr(pred, "desicion.values") > every.dv <- arrary(every.dv, dim=rep(100,3)) > contour3d(every.dv, level=0, x=every.list$a1, y=every.list$a2, z=every.lisr$a3, alpha=0.5, add=T) > plot3d(spin3d()) -------------------------------->>>> > model <- svm(x, y, cost=2, gamma=0.5, scale=F, cross=10) > plot3d(x, col=as.integer(y), type="s", radius=0.15) > plot3d(model$SV, col=as.integer(y[model$index]), type="s", radius=0.15, add=T) > pred <- predict(model, every, decision.values=T) > every.dv <- attr(pred, "decision.values") > every.dv <- array(every.dv, dim=rep(100,3)) > contour3d(every.dv, level=0, x=every.list$a1, y=every.list$a2, z=every.lisr$a3, alpha=0.5, add=T) > plot3d(spin3d()) ``` ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpb6lvxupj30lw0bkwkn.jpg) ## Presentation ### Comparative protein structure model ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpl37yc2gj30k40eswij.jpg) The steps: - Fold assignment and template slection - Target - templete alignmrnt - Model Building - Model evaluation 1. Fold assignment and template slection ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghplhevginj30ss0bsagv.jpg) Templete selection: - higher seq similarity - family of proteins - quality of template structure - Solvent, pH, ligamds... 2. Target - templete alignmrnt Once templates have been selected, a specilized method should be used to align target seq with template structures. Alignment become difficlut in "twilight zone" of less than 30% seq identity (similarity of BLUSUM62 is 62%, also ~45 & ~80). 3. Model Building ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghplnh06nwj30t00ckjym.jpg) 4. Model Evaluation Typical errors in comparative methods: - Errors in side-chain packing - Distortions and shift in correctly aligned regions - Errors in regionswithout a template - Errors due to misalignments - Incorrect template Evaluation - Having the correct fold or not - The target-template seq similarity - The environment - Having good stereochemistry on nor - Distributions if many spatial features 5. Application of comparative modeling ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpltpui1yj30tc0dqn3s.jpg) ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghplu2vnfsj30uc0c2dnd.jpg) ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghpluoezj6j30v20d047o.jpg) ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghplv3kpruj30uy0bsaif.jpg) 6. Final Conclusion ![](https://tva1.sinaimg.cn/large/007S8ZIlly1ghplvjqxz0j30sc0cqq82.jpg) **<center>Correlated** [Bioinformatics/ Introduction and Methos (Week 1 Bioinformatics Introduction)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethosweek1md) [Bioinformatics/ Introduction and Methos (Week 2 Sequence Alignment)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethosweek2md) [Bioinformatics/ Introduction and Methos (Week 3 Seq DB and BLAST Algorithnm)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek3md) [Bioinformatics/ Introduction and Methos (Week 4 Markov Model)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek4md) [Bioinformatics/ Introduction and Methos (Week 5 From Sequencing to NGS)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek5md) [Bioinformatics/ Introduction and Methos (Week6 Variant Database)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek6md) [Bioinformatics/ Introduction and Methos (Week7 Transcriptome Analysis, and RNA-Seq)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek7md) [Bioinformatics/ Introduction and Methos (Week8 Prediction and Analysis of Noncoding RNA)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek8md) [Bioinformatics/ Introduction and Methos (Week9 Ontology, Gene Ontology and KEGG Pathway Database)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek9md) [Bioinformatics/ Introduction and Methos (Week 10 Bioinformatics Database and Resources)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek10md) [Bioinformatics/ Introduction and Methos (Week 11 New Gene Evolution Detected by Genomic Computation: Basic Concepts and Examples)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek11md) [Bioinformatics/ Introduction and Methos (Week 12 From Dry to Wet, an Evolutionary Story. Evolution function analysis of DNA methyltransferase)](https://www.haoxi.info/archives/bioinformaticsintroductionandmethodsweek12md)</center>

Bioinformatics: Introduction and Methods (Week 7)

Bioinformatics: Introduction and Methods (Week 5)