![52
http://www.oncotarget.com/index.php?journal=oncotarget&page=arti
cle&op=view&path[]=23869&path[]=75083
https://www.nature.com/articles/1208329](https://image.slidesharecdn.com/may-workshop1p-180514234815/85/May-15-workshop-52-320.jpg)
May 15 workshop
- 1. 1
- 2. T-BioInfo is designed for processing, analysis and integration of multi-omics data. The platform is used in multiple research groups to extract meaningful insights from large multi-omics datasets. Our current effort expands to education, by enabling more people to extract meaningful, data-driven insights from omics datasets with biomedical applications. To learn more about the platform and it’s research and educational features, follow the highlighted links . T-bio.info | edu.t-bio.info | server.t-bio.info 2
- 3. 3
- 4. 4
- 5. 5
- 6. Modeling Precision Medicine Machine Learning forTranscriptomics Data: Extracting Meaningful insights from high-throughput biomedical data. 6
- 7. Clinical Subtypes Molecular Subtypes 7
- 8. Diagnosis, Prognosis, Response toTreatment 8 Survival prediction Treatment Selection OncotypeDXPAM50
- 9. Daemen et al., 2013, “Modeling precision treatment of breast cancer”: an analysis of over 70 different Breast Cancer cell lines and over 90 different therapeutic agents. https://genomebiology.biomedcentral.com/articles/10.1186/gb-2013-14-10-r110 9
- 10. Files we will use in this session 10
- 11. BREAK 11 Q&A
- 12. Part 1: RNA-Seq Processing from raw reads to a table of expression 12
- 13. RNA-Seq: overview .…TCTGAAACAATGCTTCAATCTAACTTATCATTCATTGGGA…. Genome 13 Gene A Gene B Gene C Transcr. ATranscript A Transcr. ATranscript C
- 14. 14 .…TCTGAAACAATGCTTCAATCTAACTTATCATTCATTGGGA….Gene A Gene B Gene C Transcr. ATranscript A Transcr. ATranscript C Reads RNA-Seq: overview
- 15. 15 .…TCTGAAACAATGCTTCAATCTAACTTATCATTCATTGGGA….Gene A Gene B Gene C Transcr. ATranscript A Transcr. ATranscript C Reads RNA-Seq: some details 1. Shattering 2. Adapters ligation 3. PCR amplification 4. “Reading”
- 16. Preprocessing: • Adapters removal plus additional • Removing PCR duplicates 16 Quantification of expression levels Mapping • Mapping on the set of known transcripts • Mapping on genome (and potential identification of novel transcripts) • Combined strategy RNA-Seq: overview
- 18. 18 Data Processing Practice Create a pipeline: 1. Upload same SVL files 2. pre-processing steps:Trimmomatic, PCRclean 3. Mapping on Genome: HiSat2 4. IsoformConstruction: Cufflinks 5. GTF Merging: Cuffmerge 6. Mapping onTranscripts: Bowtie2-t 7. Quantification: RSEMExpTable
- 20. 20 ExpressionTable Sample Name Gene ID What is this number?
- 21. Standard Measures of RNA Quantification: • Counts • FPKM – fragments per kilobase per million mapped reads: Number of reads mapped on the gene ((total number of mapped reads – in millions) x (gene length in kilobases)) • TPM – transcripts per million For one sample TPMg = C x FPKMg, where C is selected in such a way that sum of all million. Constants C are different for different samples. 21
- 22. Linear scale vs Log-scale Relative differences are biologically more meaningful than absolute. are simplified if a log-scaling is performed: Log-scaled measure = log2 (linear-scale measure + shift) For relatively large values: difference equal to 1 in log-scale is a 2x difference in linear scale; difference equal to 3 in log-scale is a 8x difference in linear scale. etc; difference equal to -1 in log-scale is a 2x difference in linear scale, but in the opposite direction. 22
- 23. Preprocessing: • Adapters removal plus additional • Removing PCR duplicates 23 Quantification of expression levels Mapping • Mapping on the set of known transcripts • Mapping on genome (and potential identification of novel transcripts) • Combined strategy RNA-Seq: overview
- 24. Comparison: the role of preprocessing 24 High expression can be affected by pre-processing steps like PCR-clean and “Trimmomatic”
- 25. BREAK 25 Q&A
- 26. BREAK 26 Q&A Error Correction – CORAL, ECHO, RACER, eMER Different Mappers – HiSat,TopHat, STAR, BWA Other Sections: • Differential Expression – CuffDiff, EDGER, DESEQ • Segmentation - BinS
- 27. Part 2: Machine Learning Data exploration and classification 27
- 30. Unsupervised analysis: PCA 30 • Explore data • Visualize Why use Principal Component Analysis? • Data Filtering • Outliers • Interpretation Considerations:
- 32. 32 Unsupervised analysis: PCA PCA 7,000 genes PCA PAM50 (35) genes Normal-like Basal Claudin-low Luminal
- 33. 33 Unsupervised analysis: Hierarchical Clustering • Identify groups • Associate sample to group Why use clustering? • Various methods • Random selection in some methods • Interpretation Considerations:
- 34. 34 Unsupervised analysis: Hierarchical Clustering
- 35. Unsupervised analysis: hierarchical clustering Dendrogram 35 2 clusters 4 clusters 8 clusters
- 36. 36 Unsupervised Analysis Practice • Remove sample IDs • Mark Group Names as ID • Run H-clust CellLines_ExprData_marked.txt
- 37. BREAK 37 Q&A
- 38. 38 DogsCats ????? Training Set Test Set Supervised Machine Learning
- 39. 39 Step-wise Linear Discriminant Analysis (swLDA)
- 40. 40 SupportVector Machine (SVM) with Linear Kernel d d
- 41. 41 SupportVector Machine (SVM) with Linear Kernel ?
- 42. ? 42 Support Vector Machine (SVM) with Linear Kernel
- 43. • Fitting classifier on training set and predicting classes on the test set • Is it possible to tune 7000 coefficients by 52 samples? • Some algorithms do feature selection: swLDA, random forest • Other algorithms won’t work if number of features >> number of samples • Curse of dimensionality 43 Considerations Supervised analysis
- 44. 44 • Extracting 15 highly informative genes from the swLDA classifier • How other supervised learning algorithms can be applied (e.g., SVM) • Feature selection can also improve quality of unsupervised learning analysis Step-wise Linear Discriminant Analysis (swLDA)
- 45. 45 Classification Practice • Organize the table with 15 genes by sample type • Color expression (green – low; red – high) • Which genes stand out? • Which sample stand out? • What groups are hard to detect? CellLines_15Genes_market.txt
- 46. 46 Classification Practice: PCA of 15 gene table
- 47. 47 Hierarchical Clustering of 15 gene table N-like Basal C-low Luminal4 clusters
- 48. BREAK 48 Q&A
- 49. Part 3: Interpretation Annotating and Interpreting Gene Expression 49
- 50. Gene annotation: ENSG to Gene Symbols plus GO 50
- 53. BREAK 53 Q&A
- 54. 1. PCA plot using top 15 genes from differential expression analysis 54 Homework:
- 55. Separation of samples from various sources:TCGA and PDX 55 2. New Datasets
- 56. 56 Part 1: Conventional Machine Learning Approaches for Next Generation Sequencing Rapid RNA-seq processing for expression quantification applying logical pipeline construction and pre-processing considerations. hands-on exercises, participants will explore the expression using conventional unsupervised machine learning methods and supervised classifiers with and without feature extraction. Using BioInfo platform, participants will learn about the logic and considerations of applying such methods and be prepared for independent downstream analysis and visualization of data downloaded R scripts produced by the system. The produced/downloaded code will be reviewed, customized and subsequent session. T-bio.info | edu.t-bio.info (FREE) | server.t-bio.info (14 days DEMO)
- 57. 57
- 58. 58 Required installations: R >= 3.4 R Studio gplots ggfortify ggplot2 ggpubr e1071 mda MASS klaR Part 2: Combining custom software with R to streamline analysis workflows and visualize ‘Omics data insights. Differential Gene Expression, Gene Set Enrichment Analysis R visualization from scratch: utilize the same dataset for basic data exploration and visualization in R. This session will strengthen the participants ability to transition to script-based workflows in RNA-seq downstream analysis and visualization. Participants will learn about downstream capabilities of R-based workflow to transform and manipulate tables and visualize findings in a meaningful way.
- 59. 59 Download and Modify R Scripts
- 60. 60
- 61. Differential expression analysis Quantities related to the degree of differential expression: • Difference between mean expression levels – fold change (please, pay attention to scale); • Statistical significance – p-value, adjusted p-value (e.g., FDR) • Level of Expression (caution with low-expressed genes from the analysis) 61
- 62. • Hard to interpret when number of groups is greater than two, so we can use Claudin-low vs normal- like groups. • Differential Expression is a natural and easy to interpret feature selection procedure. • Pathway enrichment analysis can be applied to the resulting table 62 Differential expression analysis
- 65. Gene set / pathway enrichment analysis GAGE - • Use only lists (thresholding required): one of the standard tools here isThe Database for Annotation,Visualization and Integrated Discovery – DAVID (https://david.ncifcrf.gov/home.jsp, https://david-d.ncifcrf.gov/). • Takes into consideration level of differential expression 65
- 66. 66 Gene set / pathway enrichment analysis
- 67. 67 Gene set / pathway enrichment analysis
- 68. 68 Gene set / pathway enrichment analysis Regulation of Actin Cytoskeleton B Cell Receptor Signaling Pathway
- 69. 69 Required installations: R >= 3.4 R Studio gplots ggfortify ggplot2 ggpubr e1071 mda MASS klaR Part 2: Combining custom software with R to streamline analysis workflows and visualize ‘Omics data insights. Differential Gene Expression, Gene Set Enrichment Analysis R visualization from scratch: utilize the same dataset for basic data exploration and visualization in R. This session will strengthen the participants ability to transition to script-based workflows in RNA-seq downstream analysis and visualization. Participants will learn about downstream capabilities of R-based workflow to transform and manipulate tables and visualize findings in a meaningful way.
- 70. 70 R Studio
- 71. 71