CancerVar & OPAI

Clinical interpretation of Cancer somatic Variants (CancerVar) and Oncogenic Prioritization by Artificial Intelligence (OPAI)

HOW DOES IT WORK

CancerVar takes either pre-annotated files, or unannotated input files in VCF format or ANNOVAR input format, where each line corresponds to one genetic variant; CancerVar will call ANNOVAR to generate necessary annotations. In the output, based on all 12 pieces of evidence, each variant will be assigned as "Tier_I_strong", "Tier_II_potential", "Tier_IV_benign" and "Tier_III_Uncertain" by rules specified in the AMP/ASCO/CAP 2017 guidelines.

OPAI takes 12 clinical evidence scores from CancerVar and 23 pre-computed in silico scores predicted by other computational tools from ANNOVAR as input, and predicts oncogenicity by a semi-supervised deep-learning model.

CanverVar and OPAI are Python based scripts. The user need to run CancerVar firstly as step 1 to get clinical evidence-based interpretation results and then run OPAI as step 2 if they want to get the deep-learning model-based oncogenicity prediction.

CancerVar(step 1)

SYNOPSIS

CancerVar.py [options]

WHAT DOES IT DO

CanverVar is a python script for cancer variant interpretation of clinical significance.

PREREQUISITE

1. You need install Python >=3.6
2. You need install [ANNOVAR](http://annovar.openbioinformatics.org/en/latest/) version >= 2016-02-01.
3. Most of the datases can be downloaded automatically.
4. Some updated datasets(cosmic and icgc) for Annovar: https://cancervar.wglab.org/databases/ (download and gunzip, put in the Annovar db folder)
5. Please use the updated files, outdated files will bring some problems of running CancerVar.

OPTIONS of CancerVar script

• -h, --help show this help message and exit

• --version show program''s version number and exit

• --config=config.ini Load your config file. The config file contains all options.

if you use this options,you can ignore all the other options bellow.

• -i INPUTFILE, --input=INPUTFILE input file of variants for analysis

• --input_type=AVinput The input file type, it can be AVinput(Annovar''sformat),VCF

• --cancer_type=CANCER The cancer type, please check the help for the details of cancer type: Adrenal_Gland Bile_Duct Bladder Blood Bone Bone_Marrow Brain Breast Cancer_all Cervix Colorectal Esophagus Eye Head_and_Neck Inflammatory Intrahepatic Kidney Liver Lung Lymph_Nodes Nervous_System Other Ovary Pancreas Pleura Prostate Skin Soft_Tissue Stomach Testis Thymus Thyroid Uterus,if you are using avinput file, you can can specify the cancer type in the 6th column

• -o OUTPUTFILE, --output=OUTPUTFILE prefix the output file (default:output)

• -b BUILDVER, --buildver=BUILDVER version of reference genome: hg38, hg19(default)

  CancerVar Other Options:

• -t cancervardb, --database_intervar=cancervardb The database location/dir for the CancerVar dataset files

• -s your_evidence_file, --evidence_file=your_evidence_file User specified Evidence file for each variant

  Annovar Options( check these options from manual of Annovar):

• --table_annovar=./table_annovar.pl The Annovar perl script of table_annovar.pl

• --convert2annovar=./convert2annovar.pl The Annovar perl script of convert2annovar.pl

• --annotate_variation=./annotate_variation.pl The Annovar perl script of annotate_variation.pl

• -d humandb, --database_locat=humandb The database location/dir for the Annovar annotation datasets

EXAMPLE of CancerVar

    python3.6 ./CancerVar.py -c config.ini  # Run the examples in config.ini
    python3.6 ./CancerVar.py  -b hg19 -i your_input  --input_type=VCF  -o your_output
    python3.6 ./CancerVar.py  -b hg19 -i example/FDA_hg19.av -o example/FDA

The clinical interpretation results are in the ouput file of *".cancervar", the column of "CancerVar: CancerVar and Evidence"** is the evidence and final interpretation.

OPAI(step 2)

After running CancerVar correctly and getting the output files of ".cancervar" and ".grl_p",we are ready to run Oncogenic Prioritization by Artificial Intelligence.

WHAT AND HOW DOES IT DO

OPAI is a python script for Oncogenic Prioritization by Artificial Intelligence after CancerVar. OPAI firstly call feature_preprocess.py to process the features coding from CancerVar and Annovar output, then call opai_predictor.py to predict the oncogenicity.

The OPAI scripts are in the scripts folder of “OPAI”:

• feature_preprocess.py:
  • preprocessing the ANNOVAR data and CancerVar output to generate OPAI input;
• opai_predictor.py:
  • predicting the oncogenicity of a variant.

PREREQUISITE

OPAI has currently only been tested with Python 3.6+, and requires four Python modules to be installed and in path. These are numpy https://numpy.org, pandas https://pandas.pydata.org , scikit-learn https://scikit-learn.org and pytorch https://pytorch.org.

There are two ways to install these modules:

• Using CONDA and manage the environment.

   conda create  -n opai python=3.6
   conda activate opai
   conda install -c anaconda numpy pandas scikit-learn
   conda install -c pytorch pytorch=1.9

• Using pip

  python3.6 -m pip install numpy --user
  python3.6 -m pip install pandas --user
  python3.6 -m pip install scikit-learn --user
  python3.6 -m pip install torch --user

  MODELS

  There are two trained models for prediction in OPAI, located in the folder of "saves":

• Ensemble-based model:

  • both clinical evidence score and 23 pre-computed in silico scores are taken as input of the model;
  • model file: ensemble.pt

• Evidence-based model:

  • only clinical evidence score are taken as input of the model, this is useful for case of a lot or even all the missing values in 23 pre-computed in silico scores.
  • model file: evs.pt

  Users can specify the model by using the -m ensemble or -m evs option and then following the -d model_file_location option.

  EXAMPLE of OPAI

  After running of python3.6 ./CancerVar.py -b hg19 -i example/FDA_hg19.av -o example/FDA, check files of example/FDA.hg19_multianno.txt.grl_p and example/FDA.hg19_multianno.txt.cancervar, see if they are generated correctly.

  Then,

  • using Ensemble-based model

    
    python3.6 OPAI/scripts/feature_preprocess.py -a example/FDA.hg19_multianno.txt.grl_p -c  example/FDA.hg19_multianno.txt.cancervar -m ensemble -n 5 -d OPAI/saves/nonmissing_db.npy -o example/FDA.hg19_multianno.txt.cancervar.ensemble.csv

  python3.6 OPAI/scripts/opai_predictor.py -i example/FDA.hg19_multianno.txt.cancervar.ensemble.csv -m ensemble -c OPAI/saves/ensemble.pt -d cpu -v example/FDA.hg19_multianno.txt.cancervar -o example/FDA.hg19_multianno.txt.cancervar.ensemble.pred

  
  The predicted oncogenicity are in the (last)column of **"ensemble_score"** in file `example/FDA.hg19_multianno.txt.cancervar.ensemble.pred`.

• using Evidence-based model

  
  python3.6 OPAI/scripts/feature_preprocess.py -a example/FDA.hg19_multianno.txt.grl_p -c  example/FDA.hg19_multianno.txt.cancervar -m evs -n 5 -d OPAI/saves/nonmissing_db.npy -o example/FDA.hg19_multianno.txt.cancervar.evs.csv
  
  python3.6 OPAI/scripts/opai_predictor.py -i  example/FDA.hg19_multianno.txt.cancervar.evs.csv -m evs -c OPAI/saves/evs.pt -d cpu -v example/FDA.hg19_multianno.txt.cancervar -o example/FDA.hg19_multianno.txt.cancervar.evs.pred

The predicted oncogenicity are in the (last)column of **"evs_score"** in file `example/FDA.hg19_multianno.txt.cancervar.evs.pred`.

#### OPTIONS OF OPAI SCRIPTS 
- Feature process using `feature_preprocess.py`
```bash
python3.6  OPAI/scripts/feature_preprocess.py -h
usage: feature_preprocess.py [-h] -a ANNOVAR_PATH -c CANCERVAR_PATH [-m METHOD] [-n MISSING_COUNT] -d DATABASE -o OUTPUT

feature creator from cancervar output

optional arguments:
  -h, --help            show this help message and exit
  -a ANNOVAR_PATH, --annovar_path ANNOVAR_PATH
                        the path to annovar file
  -c CANCERVAR_PATH, --cancervar_path CANCERVAR_PATH
                        the path to cancervar file
  -m METHOD, --method METHOD
                        output evs features or ensemble features (option: evs, ensemble)
  -n MISSING_COUNT, --missing_count MISSING_COUNT
                        variant with more than N missing features will be discarded, (default: 5)
  -d DATABASE, --database DATABASE
                        database for feature normalization
  -o OUTPUT, --output OUTPUT
                        the path to output

• Prediction using opai_predictor.py

  
  python3.6 OPAI/scripts/opai_predictor.py -h
  usage: opai_predictor.py [-h] -i INPUT -v CANCERVAR_PATH [-m METHOD] [-d DEVICE] -c CONFIG -o OUTPUT

optional arguments: -h, --help show this help message and exit -i INPUT, --input INPUT the path to input feature -v CANCERVAR_PATH, --cancervar_path CANCERVAR_PATH the path to cancervar file -m METHOD, --method METHOD use evs features or ensemble features (option: evs, ensemble) -d DEVICE, --device DEVICE device used for dl-based predicting (option: cpu, cuda) -c CONFIG, --config CONFIG the path to trained model file -o OUTPUT, --output OUTPUT the path to output

## Web server
We also developed a web server [http://cancervar.wglab.org](http://cancervar.wglab.org), which offers a graphical user interface for CancerVar and OPAI scores. 

This web server provided pre-compiled 13M mutations annotation results and OPAI scores. Users can directly search their exonic variants by chromosomal position, by dbSNP identifier, or by gene name with the nucleic acid/amino acid change. The web server will provide full details on the variants, including all automatically generated criteria, most of the supportive evidence and also OPAI scores.

## LICENSE

CancerVar and OPAI is free for non-commercial use without warranty. Users need to obtain licenses such as ANNOVAR by themselves. Please contact the authors for commercial use.

## REFERENCE

Quan Li, Zilin Ren, Kajia Cao, Marilyn M. Li, Yunyun Zhou and Kai Wang. CancerVar: an Artificial Intelligence empowered platform for clinical interpretation of somatic mutations in cancer ( Science Advances, 2022, [https://www.science.org/doi/10.1126/sciadv.abj1624](https://www.science.org/doi/10.1126/sciadv.abj1624) )

Quan Li and Kai Wang. InterVar: Clinical interpretation of genetic variants by ACMG-AMP 2015 guideline. The American Journal of Human Genetics 100, 1-14, February 2, 2017,[http://dx.doi.org/10.1016/j.ajhg.2017.01.004](http://dx.doi.org/10.1016/j.ajhg.2017.01.004)

[The  AMP/ASCO/CAP 2017 guidelines ](https://www.ncbi.nlm.nih.gov/pubmed/27993330)
Li MM, Datto M, Duncavage EJ, Kulkarni S, Lindeman NI, Roy S, Tsimberidou AM, Vnencak-Jones CL, Wolff DJ, Younes A, Nikiforova MN.
Standards and Guidelines for the Interpretation and Reporting of Sequence Variants in Cancer: A Joint Consensus Recommendation of the Association for Molecular Pathology, American Society of Clinical Oncology, and College of American Pathologists.

[The  ACMG/CGC 2019 guidelines ](https://www.ncbi.nlm.nih.gov/pubmed/31138931)
Mikhail FM, et al. Technical laboratory standards for interpretation and reporting of acquired copy-number abnormalities and copy-neutral loss of heterozygosity in neoplastic disorders: a joint consensus recommendation from the American College of Medical Genetics and Genomics (ACMG) and the Cancer Genomics Consortium (CGC). Genet Med. 2019 Sep;21(9):1903-1916. doi: 10.1038/s41436-019-0545-7.

## Acknowledges

Thanks to all who provided bug reports.