SureTypeSC

SureTypeSC is implementation of algorithm for regenotyping of single cell data.

Getting Started

After unpacking the installation package, run the setup.py script by supplying command below

pip setup.py install 

Prerequisites

• buildins
• git-lfs https://git-lfs.github.com/
• python 3 (tested on Python 3.8.5)
• scikit-learn = v0.23.2 (http://scikit-learn.org/stable/)
• numpy >= v1.19.2 (http://www.numpy.org/)
• pandas >= v1.1.2 (https://pandas.pydata.org/)
• IlluminaBeadArrayFiles >= 1.2.0
• json_tricks >= v3.15.3 (https://pypi.org/project/json_tricks/)
• tabulate >= v0.8.7(https://pypi.org/project/tabulate/)
• scipy==1.5.2

Usage

• create genome studio file (include name,chromosome,position, genotype, gencall score, x raw intensities, x normalized intensities, y raw instensities and y normalized intensities) [format, pandas dataframe]

import SureTypeSC as sc

df = sc.basic("HumanKaryomap-12v1_A.bpm","HumanKaryomap-12v1_A.egt","Samplesheetr.csv")


Parameters:The path of manifest file, the path of cluster file, the path of samplesheet file; 
The template of sample sheet is in Samplesheetr.csv

• return call rate of all samples over all chromosomes

import SureTypeSC.genome_library as gl

call_rate = gl.callrate(df,th=0.01)

call_rate is the data frame that includes the call rate over all the chromosomes

Parameters: df is the pandas data frame from basic function, th is the threshold based on the GenCall score.

• return call rate of all samples of one specific chromosome

call_rate_chrom = gl.callrate_chr(df,chr_name,th=0.01)

call_rate_chrom is the data frame that includes the call rate of one chromosome

Paramters: df is the pandas data frame from basic function; chr_name is the name of selected chromosome ('X'); th is the threshold based on the GenCall score

• return the M and A features of one locus

nc, ab, aa, bb = gl.locus_ma(df,locus_name)

nc, ab, aa and bb are pandas data frame with m and a features

Parameters: df is the pandas data frame from basic function; locus_name is the name of one specific locus ('rs3128117')

• return the M and A features of one chromsome of one sample

AM = gl.sample_ma(df,sample_name,chr_name)

A and M features of one chromsome of one sample

Parameters: df is the pandas data frame from basic function; sample_name is the name of the sample; chr_name is the name one chromosome

• return pca components of all samples over all chromosomes

pcs = gl.pca_samples(df,th=0.01,n=2)

pca is the pandas data frame that includes the first component and the second component.

Parameters: df is the pandas data frame from basic function; th is the threshold based on the GenCall score; n is the number of components

• return pca components of all samples of one specific chromosome

pcs_chr = gl.pca_chr(df,chr_name,th=0.01,n=2)

pcs_chr is the data frame that includes the first component and the second component of one chromosome

Parameters: df is the pandas data frame from basic function; chr_name is the name of the selective chromosome; th is the threshold based on the GenCall score; n is the number of components

• Index rearrangement (set index levels (including name chromosome and position))

dfs = sc.Data.create_from_frame(df)

dfs is Data type

• The attribute of Data type

dfs.restrict_chromosomes(['1','2']) (The parameters should be a list include the chromosome name)

dfs.apply_NC_threshold_3(0.01,inplace = True) (where 0.01 is the GenCall threshold)

• M,A calculation

dfs.calculate_transformations_2()

• Load classifier

from SureTypeSC import loader

clf = loader('clf_30trees_7228_ratio1_lightweight.clf')

clf_2 = loader('clf_GDA_7228_ratio1_58cells.clf') (input should be the path of classifier)

• predict

result_rf = clf.predict_decorate(dfs,clftype='rf',inn=['m','a'])  (test is the dataset,clftype is the short for classifier like 'rf' or 'gda'. inn is the input feature)

result_gda = clf.predict_decorate(result_rf,clftype='gda',inn=['m','a'])

• Train and predict

train = sc.Trainer(result_rf,clfname='gda',inner=['m','a'],outer='rf_ratio:1.0_pred')

train.train()

result_end = train.predict_decorate(result_gda,clftype='rf-gda',inn=['m','a'])

• save the result

result_end.save_complete_table('fulltable.txt',header=False)

• save the different modes

recall mode: result_end.save_mode('recall','recall.txt',header=False,ratio=1.0)

standard mode: result_end.save_mode('standard','st.txt',header=False,ratio=1.0)

precision mode: result_end.save_mode('precision','precision.txt',header=False,ratio=1.0)

customized saving: result_end.scsave('name.txt', header=True, clftype='rf',threshold=0.15)

The program enriches every sample in the input data by :

| Subcolumn name  | Meaning |
| ------------- | ------------- |
| rf_ratio:1_pred  | Random Forest prediction (binary)  |
| rf_ratio:1_prob  | Random Forest Score for the positive class |
| gda_ratio:1_prob | Gaussian Discriminant Analysis score for the positive class  | 
| gda_ratio:1_pred | Gaussian Disciminant Analysis prediction (binary) | 
| rf-gda_ratio:1_prob | combined 2-layer RF and GDA - probability score for the positive class | 
| rf-gda_ratio:1_pred | binary prediction of RF-GDA | 

Contact

In case of any questions please contact Ivan Vogel (ivogel@sund.ku.dk)