1. Johns Hopkins University
ASHG Workshop Classifying and Interpreting Germline and Somatic Variants in Your Large Cohort Study with the CRAVAT Tool Suite
Karchin Lab
Department of Biomedical Engineering Institute of Computational Medicine
Johns Hopkins University
WiFi
SSID: ASHGWORKSHOP
Password: ASHGWORKSHOP

2. Wi-Fi Information: Hilton Ballrooms
Select the SSID: ASHGWORKSHOP
Enter password: ASHGWORKSHOP
Password is case-sensitive

3. Disclosure for:
ASHG Interactive Workshop: Classifying and Interpreting Germline and Somatic Variants in Your Large Cohort Study with the CRAVAT Tool Suite
No Relevant Conflicts to Disclose:
Rachel Karchin
Michael Ryan

4. CRAVAT/MuPIT Workshop
Overview
CRAVAT Login, Job Submission, Results Viewer
Study Overview
Variant Analysis
Impact
What change is a variant introducing?
Importance
How to sort/filter to identify variants of interest?
Investigate
What annotations can be used to further investigate variant relevance?
MuPIT – 3D Structure analysis of Variants

5. Variant Studies
High volume sequencing delivers hundreds of called variants across hundreds of samples.
Many genomic annotation packages (e.g. Annovar) provide mapping of variants onto transcripts and proteins, and provide annotations and scores.
But: Identifying interesting mutations and exploring the impact of the mutations remains challenging and often requires bioinformatics specialists.
CRAVAT / MuPIT were developed to meet this need for users without bioinformatics/biostatistics experience using machine learning and a highly visual, dynamic interface.

6. The number of variants detected is large and is getting larger
70M SNVs from WGS
638 patients
967 controls
>70M exonic germline variants!
Dr. Nicholas Roberts Department of Pathology
Johns Hopkins Medicine

7. CRAVAT Cancer-Related Analysis of VAriants Toolkit CRAVAT
Masica DL, Douville C, Tokheim C, Bhattacharya R, Kim R, Moad K, Ryan MC, Karchin R (2017). CRAVAT: Cancer-Related Analysis of VAriants Toolkit. CRAVAT 4: Cancer-Related Analysis of Variants Toolkit. Cancer Research. [in press].
CHASM
Carter H, Chen S, Isik L, Tyekucheva S, Velculescu VE, Kinzler KW, Vogelstein B, Karchin R. Cancer-specific high-throughput annotation of somatic mutations: computational prediction of driver missense mutations Cancer Res Aug 15;69(16):
SNVBox
Wong WC, Kim D, Carter H, Diekhans M, Ryan M, Karchin R (2011). CHASM and SNVBox: toolkit for detecting biologically important single nucleotide mutations in cancer. Bioinformatics, 27(15):
VEST
Carter H, Douville C, Stenson P, Cooper D, Karchin R (2013) Identifying Mendelian disease genes with the Variant Effect Scoring Tool. BMC Genomics, 14(Suppl 3):S3.

8. CRAVAT is designed to filter and prioritize variants
Funnel for automated
variant analysis
mention MIke’s demo time and place

9. CRAVAT http://www.cravat.us
CRAVAT is a free, web-based tool for high-throughput analysis of genomic variants.
Genomic
Features
Machine
Learning
Algorithms
Parallel Processing
Server
Spreadsheet Results
Annotations
Interactive Results

10. CRAVAT Analysis – Setup
Click here to
create
an account

11. CRAVAT Analysis – Sample Job
Click here to
see jobs.

12. CRAVAT Analysis – Sample Job
Click here for
Interactive
Results Viewer

13. CRAVAT Analysis – Sample Job Interactive Results
Tabs for
different
lists
Click header
to sort
Select row to
see details
Type to
Filter
Add/remove
columns

14. CRAVAT Input Format
Format 1 – Simple Cravat Format (tab separated) ID Chr Position Strand RefBase AltBase Single Base Substitution: Mut1 chr C T Insert: Mut2 chr G Delete: Mut3 chr G - Multi-base Insert: TR4 chr AGG

15. CRAVAT Input Format http://projects.insilico.us.com/CRAVATClass
Exercise 1:
Use the link on the site above to get Exercise1.txt.
Correct format errors. (Use CRAVAT Help)
Copy / Paste into input screen and submit
In the results, can you find all 9 variants that were input?

16. CRAVAT – Input Format
Exercise 1: Can you find all 9 variants that were input?
Tip: In the Columns / Variant Info add Input Line and ID to help find particular variants.

17. CRAVAT Input Format
The optional sample ID column can be used when providing study data for a cohort:
Mut1 chr T C Patient1
Mut2 chr T C Patient1
Mut5 chr T G Patient1 …
Mut234 chr T G Patient5
Mut789 chr T G Patient11
Mut1036 chr T G Patient21
. .

18. VCF Input Format CRAVAT Input Format – VCF File Exercise 2:
Download Exercise2.vcf from the class webpage. Submit the file to CRAVAT as a new job.
How may input variants are in the file?
How many input results did you receive from CRAVAT?
Add VCF Specific “Mutation Call Quality” columns
What additional information do these fields provide?

19. CRAVAT – Input Format VCF
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT NA NA NA00003
TR1 A T 29 PASS NS=3;DP=14;AF=0.5;DB;H2 GT:GQ:DP:HQ 0|0:48:1:51,51 1|0:48:8:51,51 1/1:43:5:.,.
GGAAGAAGAA G,GGAA,GGAAGAA 50 PASS NS=3;DP=13;AA=T GT:GQ:DP:HQ:AD 0|1:48:4:51,51:1,1,1,1 1|2:21:6:23,27:1,2,1,2 3|1:48:8:51,51:0,3,3,2

20. CRAVAT – Input VCF

21. CRAVAT – Results Overview
Exercise 3:
Open BRCA Study Results – random cohort of TCGA somatic breast cancer variants. Link on class page:
Use the Summary Tab:
What percentage of study variants are coding?
What gene function GO category is most heavily mutated?
What percentage of study variants are missense mutations if you discard synonymous variants?
Which chromosome has a high density of inactivating mutations? Which genes occur in those locations?
Which genes are pervasively mutated in study participants?
Which patients had the most / least variants. Can you graph stopgain mutations by patient?
Which gene pathways are enriched for likely cancer driver variants as predicted by CHASM score? Can you visualize the pathway and see the mutated genes?

22. CRAVAT – Results Overview

23. CRAVAT Analysis – Impact Analysis
CRAVAT will determine the impact of a variant (sequence ontology), identify affected genes, and determine codon/protein coding changes.
CRAVAT will evaluate variant impacts on all the transcripts of a gene.
Code
Meaning SY
Synonymous Variant SL
Stop Lost SG
Stop Gained MS
Missense Variant II
In Frame Insertion FI
Frameshift Insertion ID
In Frame Deletion FD
Frameshift Deletion SS
Splice Site

24. CRAVAT – Variant Impact
Exercise 4:
Using the same results:
Go to Variant Tab
With Columns panel, Add ID to your display
What is the variant impact: gene, sequence ontology, protein change caused by the variants with the IDs:
r r01448
r r07405
r r02152
r09656

25. CRAVAT – Variant Impact

26. CRAVAT – Variant Scoring
Scoring methods are needed to help prioritize investigation of mutations.
CRAVAT provides two scoring methods:
VEST – Machine learning algorithm trained to identify pathogenic variants. Single base variants, indels (frameshift and in frame), splice site variants, and nonsense variants (stop codon changes).
CHASM – Machine learning algorithm trained to identify cancer drivers. Missense Only
CRAVAT analysis can include VEST scores, CHASM scores or both.

27. VEST and CHASM bioinformatic variant scores
Random forest machine learning uses an ensemble of simple decision trees

28. VEST and CHASM bioinformatic variant scores
Supervised machine learning
Training
“Bad” variants
Feature encoding
Harmless variants
Random Forest
85 features
sequence-based
evolutionary conservation in protein and DNA multiple alignments
predicted protein structure
positional- and regional protein annotations

29. VEST and CHASM bioinformatic variant scores
Supervised machine learning
Prediction
Feature encoding
New variant s
0.8
85 features
sequence-based
evolutionary conservation in protein and DNA multiple alignments
predicted protein structure
positional- and regional protein annotations
Random Forest

30. VEST and CHASM use a statistical model to threshold variant scores
Statistical significance of tree vote score?
From scores to P-values
Null hypothesis is that there are no “bad” variants present
sorting variants by scores
Null tree vote scores
Allows comparison of scores for different mutation
consequence types.

31. CHASM performance
CHASM remains one of the top-performing predictors of driver missense mutations
Benchmarking done by an independent group

32. VEST is one of the top performing predictors of missense pathogenicity
Benchmarking done by an independent group
VEST3.0
VEST3.0

33. VEST provides integrated prioritization of all non-silent variants
VEST classifiers have been built for missense, frameshift, nonsense, and splice site variants.
Scores are not directly comparable across different variant types
- But -
P-value may be directly compared across types making it easier to sort all variants by predicted pathogenic impact.
Precomputed VEST scores for all possible missense mutations have been developed so VEST scores can be quickly returned for single variant lookups and are included in other annotation databases.

34. CRAVAT – Variant Scoring
Exercise 5
Using the same results, variant tab:
Sort by CHASM p-value (lowest at top of list).
What type of variants are identified as potential cancer drivers? Why?
What genes have the variants at the top of the list? Are they familiar? Are there unexpected genes?
Sort by VEST p-value (lowest at the top of the list)
What type of variants are at the top of the list. Why?
If you wanted to see the most pathogenic frameshift deletions can you sort first by variant type and then by vest p-value?
Are VEST scores the same for all transcripts of a gene?

35. CRAVAT Analysis –Variant Scoring

36. CRAVAT – Annotations
Population Statistics / Gene Information

37. CRAVAT – Annotations Cancer / Disease annotations
CGL – Oncogene / Tumor Suppressor

38. CRAVAT – Annotation
Exercise 6
Using the same results, variant tab:
Variant r09138 has a good CHASM p-value. Use the annotation data to decide if you would further pursue it as a driver variant.
Variant r07683 also has a good CHASM score. Is there annotation data to support the prediction that it is acting as a cancer driver? Assume the variants in analysis were from a breast cancer study. What does the gene affected by the variant do?
Splice Site mutation r01466 has a good VEST score. Is there supporting evidence that this variant is pathogenic?

39. CRAVAT – Annotation

40. CRAVAT Gene Level Tab

41. Very large number of variants Variant calls that pass QC
Variant consequence types of interest
Below allele frequency of interest
Bioinformatics variant impact prediction
Annotations
splice not yet supported - in progress
TRACTABLE LIST OF THE MOST INTERESTING variants

42. Example of funnel developed for familial pancreatic cancer (FPC) WGS study
well-known FPC genes
Genetic basis of FPC unknown for 80-90% of patients
Analysis of 70M SNVs identifies new genes not previously implicated in FPC susceptibility

43. CRAVAT Analysis Miscellaneous CRAVAT Tips:
Spreadsheet and interactive results are provided.
Spreadsheet has multiple tabs.
Gene level tab shows aggregated information by gene (e.g. how many mutations in the study)
CRAVAT is in GRCh38 coordinates if you have data in hg19 coordinates, use the checkbox to ‘lift over’ the coordinates or use the older CRAVAT at hg19.cravat.us.
If you run more than 100,000 coding mutations, result filtering will be needed to see results.
If you process more than 60,000 mutations, results will be in a text file rather than spreadsheet.
Results are not kept on the CRAVAT server indefinitely. Get results before 30 days have elapsed.
FDR will not be calculated unless at least 10+ variants are scored.
Export can be used to send on screen variant list to a spreadsheet.

44. CRAVAT Advanced Interfaces
Web Service
Fast Single Mutation Lookup
Program /
Web Server
RESTful Web Service
Submit Full Asynchronous Jobs
Program
CRAVAT Web Server
CRAVAT in Galaxy
Run CRAVAT in your own
Docker Container

45. CRAVAT Coming Soon Non-coding annotations
5’ 3’ UTR, Intron, upstream/downstream
Non-coding RNA
Pseudogene
LINE, SINE, repeats
GWAS
Regulator and chromatin modification regions
Additional Mutation Databases for Lollipop and MuPIT
Genome Aggregation Database (gnomAD)
Modular CRAVAT
Install locally with just the annotation modules you want
Develop your own annotation modules and make them available to other CRAVAT users.

46. MuPIT Visualization
3D Structure Viewer with mutation mapping capabilities.
Niknafs N, Kim D, Kim R, Diekhans M, Ryan M, Stenson PD, Cooper DN, Karchin R. MuPIT interactive: webserver for mapping variant positions to annotated, interactive 3D structures. Hum Genet Nov;132(11): doi: /s Epub 2013 Jun 23. Results are not kept on the CRAVAT server indefinitely. Get results before 30 days have elapsed.
Identify 3D mutation clusters
Combine mutation with annotation to identify proximity of mutation to critical protein structures.

47. MuPIT Visualization
Exercise 7:
Using the same results, variant tab:
On r08293 in the detail panel, click on the MuPIT button to see the variant on STK3’s 3D structure.
Click and drag with your mouse to rotate the structure. Use wheel to zoom in / out.
Change the color of your mutation to red.
On the Chains section, set the color of each protein chain to a different color to better understand the structure.
Use the style menus in the Protein tab to see different renderings of the protein.
Use the small molecule buttons on the Protein tab and active/binding buttons on the annotation tab to discover potential impacts of the variant.

48. MuPIT Visualization

49. MuPIT Visualization
Exercise 8:
Using the same results, variant tab:
On r04947 in the detail panel, click on the MuPIT button to see the variant on a 3D structure.
Use the Results button to switch to the homology model of NP_ _9. Turn off your mutation. Try the TCGA variant buttons for BRCA to see other breast cancer variants for MAP3K1.
Turn off the variants but leave the hot region (right click on the BRCA button)
Turn your variant back on. What is the relationship between your study mutation and the hot region?

50. MuPIT Visualization

51. MuPIT from Gene Panel

52. MuPIT Stand-Alone

54. Principal Collaborators and Funding for the Project
Johns Hopkins University
Rachel Karchin
David Masica
Collin Tokheim
Chris Douville
Violeta Beleva
Lily Zheng
In Silico Solutions
Michael Ryan
RyangGuk Kim
Kyle Moad
Funding for CRAVAT/MuPIT
Information Technologies for Cancer Research, NCI (grants U01-CA180956, U24CA204817)