CHASMplus: predicting which missense mutations drive human cancers¶
| Author: | Collin Tokheim, Rachel Karchin |
|---|---|
| Contact: | ctokhei1 # alumni DOT jh DOT edu |
| Lab: | Karchin Lab |
| Source code: | GitHub |
| Q&A: | Biostars (tag: CHASMplus) |
Large-scale DNA sequencing studies of patients’ tumors have revealed that most driver mutations occur only in a few patients, which presents a challenge for precision medicine. CHASMplus is a machine learning method that accurately distinguishes between driver and passenger missense mutations, even for those found at low frequencies or are cancer type-specific. Unlike previous approaches that focus on identifying driver genes, CHASMplus identifies whether individual mutations are cancer drivers. CHASMplus can be used by both bioinformaticians and biolgists by using a graphical user interface or a command line tool.
Note
You can run CHASMplus without installing anything by submitting your data to the OpenCRAVAT webserver (details here). After creating a user account, you’ll just need to check the box for CHASMplus and hit the annotate button (OpenCRAVAT webserver). Also, you can install locally a graphical user interface [see the Quick start (OpenCRAVAT & CHASMplus)]
Prominent papers using CHASMplus:
- Reiter et al., Minimal functional driver gene heterogeneity among untreated metastases. Science
- Anagnostou, Niknafs et al., Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nature Cancer
- Saito et al., Landscape and function of multiple mutations within individual oncogenes. Nature
- Reiter et al., An analysis of genetic heterogeneity in untreated cancers. Nature Reviews Cancer
- Hu et al., Multi-cancer analysis of clonality and the timing of systemic spread in paired primary tumors and metastases. Nature Genetics
- Sakomoto et al., The Evolutionary Origins of Recurrent Pancreatic Cancer, Cancer Discovery
Contents:
Quick start (OpenCRAVAT & CHASMplus)¶
The easiest way to obtain CHASMplus scores is by using OpenCRAVAT to fetch precomputed scores.
Install OpenCRAVAT app¶
You can install OpenCRAVAT app for Mac or Windows through the installers provided:
Launch the OpenCRAVAT app, and when prompted install the required system-level annotators to complete the installation process.
Next, click the “Store” tab.
Search for CHASMplus in the store and then click the CHASMplus annotator. You can select either to install the default CHASMplus or a cancer type-specific version.
Next, press the “install” button.
It will take several minutes to install, depending on the speed of your internet. By going back to the “jobs” tab in the upper left, you are now ready to submit mutations!
Test example¶
Please download the example input here.
For this example we use a simple tab-delimited format, but OpenCRAVAT can also handle VCF files. The tab-delimited format consists of the following 7 columns: 1) chromosome (with “chr” prefix); 2) Start position (1-based coordinates); 3) Strand (“+” or “-“); 4) Reference allele (“-” for insertions); 5) Alternate allele (“-” for deletions); 6) sample ID; 7) variant ID. For more details about input file formats, please see the OpenCRAVAT wiki.
There are four steps to run CHASMplus: 1) upload the input.txt example file (make sure the genome is set to hg38); 2) Make sure CHASMplus is checked in the annotator section; 3) Press the Annotate button in the bottom left; 4) Click to generate an excel output file. CHASMplus results will be on the “variant” tab.
By clicking the “launch” button, you can also interactively explore the results in OpenCRAVAT. Like the excel spreadsheet, CHASMplus results are found on the “variant” tab.
While this tutorial only had you run one annotator (CHASMplus), OpenCRAVAT has 50+ more annotations available.
Interpretation¶
CHASMplus scores range from 0 to 1, with higher scores meaning more likely to be a cancer driver mutation. If you are looking to identify a discrete set of putative driver mutations, then we suggest that you correct the p-values for multiple hypothesis testing. We recommend using the Benjamini-Hochberg (BH) procedure for controling the false discovery rate. You will need to use an external package to do this, e.g., the p.adjust function in R. False discovery rate adjustments will likely be added in the future.
Further documentation¶
For further advanced features of OpenCRAVAT, please see the OpenCRAVAT wiki.
Install Command Line OpenCRAVAT¶
Note
The command line version is meant for users with bioinformatic experience
You will need python 3.6 or newer to use OpenCRAVAT. You will first need to install the OpenCRAVAT python package, please follow the instructions on the OpenCRAVAT wiki:
Install CHASMplus annotators¶
OpenCRAVAT has a modular architecture to perform genomic variant interpretation including variant impact, annotation, and scoring. CHASMplus is one module available in the CRAVAT store. To install the CHASMplus module within OpenCRAVAT, please execute the following command:
$ cravat-admin install chasmplus
The above command may take a couple minutes and will install the pan-cancer model of CHASMplus scores. To install cancer type specific versions of CHASMplus, follow the following template:
$ cravat-admin install chasmplus_LUAD
where LUAD, the abbrevitation from the The Cancer Genome Atlas, designates lung adenocarcinoma. To see a full list of available annotators, issue the following commnad:
$ cravat-admin ls -a -t annotator
Running CHASMplus¶
OpenCRAVAT takes as input either a VCF file or a simple tab-delimited text file. I will describe a simple example that uses the latter. The simple tab-delimited text file should contain a variant ID, chromosome (with “chr”), start position (1-based), strand, reference allele, alternate allele, and optional sample ID.:
chr10 122050517 + C T sample1 var1
chr11 124619643 + G A sample1 var2
chr11 47358961 + G T sample1 var3
chr11 90135669 + C T sample1 var4
chr12 106978077 + A G sample1 var5
You can download an example input file here.
Note
By default, OpenCRAVAT processes variants on the hg38 reference genome. If you are using hg19 or hg18, please specify with the “-l” parameter your specific reference genome so that OpenCRAVAT will know to lift over your variants.
You can run CHASMplus by using the cravat command. For information about command line options, please see the command line help:
$ cravat -h
To obtain CHASMplus scores for pan-cancer (annotator “chasmplus”) and lung adenocarcinoma (annotator “chasmplus_LUAD”), run the following command:
$ cravat -n MYRUN -t excel -a chasmplus chasmplus_LUAD -d output_directory input.txt
The above command will run all annotators (specified by the -a flag, multiple separated by a space) and save results to the directory named “output_directory”. The “-t” option specifies the output to be saved as an excel file. The -n flag specifies the name of the run. Scores and p-values from CHASMplus are found in the “MYRUN.xlsx” file (or “MYRUN.tsv” if -t text is chosen). You should see the “Variant” excel sheet that contains columns like this:
CHASMplus CHASMplus_LUAD
P-value Score Transcript All results P-value Score Transcript All results
0.399 0.048 ENST00000453444.6 ENST00000334433.7:(0.025:0.59),ENST00000358010.5:(0.049:0.393),*ENST00000453444.6:(0.048:0.399),NM_001291876.1:(0.046:0.412),NM_001291877.1:(0.045:0.418),NM_206861.2:(0.048:0.399),NM_206862.3:(0.025:0.59) 0.644 0.013 ENST00000334433.7 *ENST00000334433.7:(0.013:0.644),ENST00000358010.5:(0.023:0.478),ENST00000453444.6:(0.022:0.492),NM_001291876.1:(0.022:0.492),NM_001291877.1:(0.022:0.492),NM_206861.2:(0.023:0.478),NM_206862.3:(0.013:0.644)
0.99 0.001 NM_052959.2 *NM_052959.2:(0.001:0.99) 0.945 0.002 NM_052959.2 *NM_052959.2:(0.002:0.945)
0.446 0.041 NM_001080547.1 ENST00000533968.1:(0.053:0.369),*NM_001080547.1:(0.041:0.446),NM_003120.2:(0.049:0.393) 0.278 0.044 NM_001080547.1 ENST00000533968.1:(0.043:0.284),*NM_001080547.1:(0.044:0.278),NM_003120.2:(0.053:0.224)
CHASMplus scores are provided in a transcript specific manner, with the score for the default selected transcript shown in the “Score”, “P-value”, and “Transcript” columns. Scores for other transcripts are listed in the “All results” column.
Interpretation¶
CHASMplus scores range from 0 to 1, with higher scores meaning more likely to be a cancer driver mutation. If you are looking to identify a discrete set of putative driver mutations, then we suggest that you correct for multiple hypothesis testing. We recommend using the Benjamini-Hochberg (BH) procedure for controling the false discovery rate. You will need to use an external package to do this, e.g., the p.adjust function in R. False discovery rate adjustments will likely be added in the future.
Further documentation¶
For further advanced features of OpenCRAVAT, please see the OpenCRAVAT wiki.
Available CHASMplus models¶
CHASMplus can perform predictions either using a cancer type-specific model or in a “pan-cancer” manner by consider multiple cancer types together. Pan-cancer is a useful default if a matching cancer type is not available from The Cancer Genome Atlas (TCGA). We have made the following results available through OpenCRAVAT:
| Annotator name | Data source | Cancer type |
|---|---|---|
| chasmplus | TCGA | Pan-cancer (multiple cancer types) |
| chasmplus_LAML | TCGA | Acute Myeloid Leukemia |
| chasmplus_ACC | TCGA | Adrenocortical carcinoma |
| chasmplus_BLCA | TCGA | Bladder Urothelial Carcinoma |
| chasmplus_LGG | TCGA | Brain Lower Grade Glioma |
| chasmplus_BRCA | TCGA | Breast invasive carcinoma |
| chasmplus_CESC | TCGA | Cervical squamous cell carcinoma and endocervical adenocarcinoma |
| chasmplus_CHOL | TCGA | Cholangiocarcinoma |
| chasmplus_COAD | TCGA | Colon adenocarcinoma |
| chasmplus_ESCA | TCGA | Esophageal carcinoma |
| chasmplus_GBM | TCGA | Glioblastoma multiforme |
| chasmplus_HNSC | TCGA | Head and Neck squamous cell carcinoma |
| chasmplus_KICH | TCGA | Kidney Chromophobe |
| chasmplus_KIRC | TCGA | Kidney renal clear cell carcinoma |
| chasmplus_KIRP | TCGA | Kidney renal papillary cell carcinoma |
| chasmplus_LIHC | TCGA | Liver hepatocellular carcinoma |
| chasmplus_LUAD | TCGA | Lung adenocarcinoma |
| chasmplus_LUSC | TCGA | Lung squamous cell carcinoma |
| chasmplus_DLBC | TCGA | Lymphoid Neoplasm Diffuse Large B-cell Lymphoma |
| chasmplus_MESO | TCGA | Mesothelioma |
| chasmplus_OV | TCGA | Ovarian serous cystadenocarcinoma |
| chasmplus_PAAD | TCGA | Pancreatic adenocarcinoma |
| chasmplus_PCPG | TCGA | Pheochromocytoma and Paraganglioma |
| chasmplus_PRAD | TCGA | Prostate adenocarcinoma |
| chasmplus_READ | TCGA | Rectum adenocarcinoma |
| chasmplus_SARC | TCGA | Sarcoma |
| chasmplus_SKCM | TCGA | Skin Cutaneous Melanoma |
| chasmplus_STAD | TCGA | Stomach adenocarcinoma |
| chasmplus_TGCT | TCGA | Testicular Germ Cell Tumors |
| chasmplus_THYM | TCGA | Thymoma |
| chasmplus_THCA | TCGA | Thyroid carcinoma |
| chasmplus_UCS | TCGA | Uterine Carcinosarcoma |
| chasmplus_UCEC | TCGA | Uterine Corpus Endometrial Carcinoma |
| chasmplus_UVM | TCGA | Uveal Melanoma |
Advanced: download (source)¶
CHASMplus releases¶
- CHASMplus v1.0.0 - 8/17/2018 - Initial release
Necessary additional code¶
- 20/20+ code produces driver gene scores. Please follow installation instructions from the 20/20+ website.
- SNVBox code fetches the features used by CHASMplus from a MySQL database
Necessary data files¶
- SNVBox MySQL database
- Pre-computed scores data set
- Reference SNVBox transcripts in BED format
Advanced: Build from Source¶
CHASMplus is only intended to be ran on linux operating systems and on a compute server.
Package requirements¶
CHASMplus Environment¶
We recommend using conda to install the CHASMplus dependencies.
$ conda env create -f environment.yml # create environment for CHASMplus
$ source activate CHASMplus # activate environment for CHASMplus
Make sure the CHASMplus environment is activated when you want to run CHASMplus.
20/20+¶
You will need to download the 2020plus github repository. Please follow the installation instructions from the 20/20+ website.
Set the directory of 20/20+ in the configuration file for CHASMplus. You can find this configuration file within the CHASMplus directory at chasm2/data/config.yaml.
twentyTwentyPlus: /path/to/2020plus # set this directory
Check your PATH variable¶
Make sure that you have add the 20/20+ directory to your PATH variable. If you have done this correctly, the following command should print the location of the 2020plus.py script.
$ which 2020plus.py
SNVBox database (MySQL)¶
Features for mutations CHASMplus are obtained can also be prepared by directly using a MySQL database. A MySQL dump of the SNVBox database contains features used for our study. The SNVBox database has a fairly large file size, you may want to directly download and upload to MySQL.
$ wget http://karchinlab.org/data/CHASMplus/SNVBox_chasmplus.sql.gz
$ gunzip SNVBox_chasmplus.sql.gz
$ mysql [options] < SNVBox_chasm2.sql
This will create a database named mupit_modbase, where [options] is the necessary MySQL parameters to login. You will need sufficient privileges on your MySQL database to CREATE a new database. If everything worked properly, you should see a database named “SNVBox_20161028_sandbox”.
SNVBox code¶
The next step is to download the code that fetches features from the SNVBox database. Please download the code from here, or use wget:
$ wget http://karchinlab.org/data/CHASMplus/SNVBox.tar.gz
The next step is to set the configuration file (snv_box.conf) to point towards the established database in the previous section. Specifically, change the db.user, db.password, and db.host to point towards your own mysql user name, mysql password, and mysql host.
The last step is to set the CHASMplus configuration file to point towards the path of the snvGetGenomic command within the SNVBox code. The yaml configuration file is found within the CHASMplus directory at chasm2/data/config.yaml.
snvGetGenomic: /path/to/SNVBox/snvGetGenomic # set this path
FAQ¶
Who should I contact if I encounter a problem?
If you believe your problem may be encountered by other users, please post the question on biostars. Check to make sure your question has not been already answered by looking at posts with the tag CHASMplus. Otherwise, create a new post with the CHASMplus tag. We will be checking biostars for questions. You may also contact me directly at ctokhei1 AT alumni DOT jh DOT edu.
Does CHASMplus support targeted gene panels?
Yes! We have added CHASMplus modules into OpenCRAVAT that support targeted gene panel sequencing. The first iteration was done for the MSK-IMPACT gene panel, but others may be supported upon request. Please see the OpenCRAVAT instructions for how to install “annotators” (GUI, command line). The MSK-IMPACT version of CHASMplus will have “MSK-IMPACT” (GUI version) or “mski” (command line version) in the name.
Can I get custom scores based on my own data from targeted gene panels?
Yes. However, you will need to run the source code version of CHASMplus to customize the predictions to your data.
Where can I obtain the training data for CHASMplus?
You can obtain the set of mutations used for training from here.
I want to compare my method to CHASMplus. How should I do it?
I recommend using the precomputed scores available through OpenCRAVAT [see Quick start (OpenCRAVAT & CHASMplus)]. Scores in the precompute were generated using gene-hold out cross-validation, so there is no issue when evaluating performance about training set overlap leading to overfitting. However, the scores do reflect training based on data from The Cancer Genome Atlas (TCGA). If a new method is trained using more data than is available from the TCGA, then it is recommended to create a new CHASMplus model based on the larger data set by using the CHASMplus source code.
I want to apply CHASMplus to new data. How should I do it?
For small datasets it is recommended that pre-computed scores obtained from OpenCRAVAT are used. Large datasets may also use the pre-computed scores, but won’t benefit from predictions that are customized to your data. If a cancer type you are interested in does not have precomputed scores or you have collected a large number of cancer samples, than it is recommended to use the CHASMplus source code to perform tailored predictions for your data.
Citation¶
Please cite our paper:
Tokheim and Karchin, CHASMplus Reveals the Scope of Somatic Missense Mutations Driving Human Cancers, Cell Systems (2019), https://doi.org/10.1016/j.cels.2019.05.005