Easy-Prime: Machine learning based pegRNA design

Contents:

Ask questions here

https://github.com/YichaoOU/easy_prime

Summary

PE design involves carefully choosing a standard sgRNA, a RT template that contains the desired edits, a PBS that primes the RT reaction, and a ngRNA that nicks the non-edit strand. Usually thousands of combinations are available for one single disired edit. Therefore, it is overwhelming to select the most likely high-efficient candidate from the huge number of combinations.

Easy-Prime applies a machine learning model (i.e., XGboost) that learned important PE design features from public PE amplicon sequencing data to help researchers selecting the best candidate.

Installation

conda create -n genome_editing -c cheng_lab easy_prime

source activate genome_editing

easy_prime -h

For detailed installation with screenshots, see: Installation

Input

  1. vcf input example

VCF headers will be ignored. Only the first 5 columns from the vcf file will be used; they are: chr, pos, name/id, ref, alt.

## comment line, will be ignored
chr9    110184636       FIG5G_HEK293T_HEK3_6XHIS        G       GCACCATCATCACCATCAT
chr1    185056772       FIG5E_U2OS_RNF2_1CG     G       C
chr1    173878832       rs5878  T       C
chr11   22647331        FIG3C_FANCF_7AC_PE3B    T       G
chr19   10244324        EDFIG5B_DNMT1_dPAM      G       T

  1. fasta input example

To specify reference and alternative allele, you need two fasta sequences; _ref is a keyword that will be recognized as the reference allele and _alt is a keyword for target mutations.

>test_ref
AAAAAAAAAAAAAAAAAAAAAAAAAGGAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACCAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
>test_alt
AAAAAAAAAAAAAAAAAAAAAAAAAGGAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACGAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

Config file

Default values are shown in the following yaml files.

genome_fasta: /path/to/genome.fa
scaffold: GTTTTAGAGCTAGAAATAGCAAGTTAAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTCGGTGC
debug: 0
n_jobs: 4
min_PBS_length: 8
max_PBS_length: 17
min_RTT_length: 10
max_RTT_length: 25
min_distance_RTT5: 3
max_ngRNA_distance: 100
max_target_to_sgRNA: 10
sgRNA_length: 20
offset: -3
PAM: NGG

Output

The output folder contains:

  • topX_pegRNAs.csv

  • rawX_pegRNAs.csv.gz

  • X_p_pegRNAs.csv.gz

  • summary.csv

The top candidates are provided in topX_pegRNAs.csv. This is a rawX format file.

rawX format

X means the input to machine learning models. Here, rawX basically means the file before machine learning featurization. Specifically, rawX contains 11 + 1 columns. The first 5 columns are from the input vcf file: sample_ID, chr, pos, ref, alt, where sample_ID ends with _candidate_xxx, this indicates the N-th combination. The next 6 columns are genomic coordinates: type, seq, chr, start, end, strand, where the type could be sgRNA, PBS, RTT, or ngRNA. Since for one PE design, it has to have these 4 components, which means that for one unique sample_ID, it has 4 rows specifying the sequences for each of them. The 12-th column, which is optional, is the predicted efficiency; in other words, the Y for machine learning.

Both topX_pegRNAs.csv and rawX_pegRNAs.csv.gz use this format.

X format

X format is the numeric representation of rawX. X_p format appends the predicted efficiency to the last column of X.

Main results

The main results, which is the top condidates, is provided in topX_pegRNAs.csv.

PE design visualization

Users can visualize the predicted combinations using:

easy_prime_vis -f topX_pegRNAs.csv -s /path/to/genome_fasta.fa

This will output pdf files to a result dir.

Usage

git clone https://github.com/YichaoOU/easy_prime

cd easy_prime/test

easy_prime -h

easy_prime --version

## Please update the genome_fasta in config.yaml

easy_prime -c config.yaml -f test.vcf

## Will output results to a folder

DASH application

Easy-Prime also provides a dash application.

Please have dash installed before running the dash application.

git clone https://github.com/YichaoOU/easy_prime

cd easy_prime/dash_app

python main.py