Methods of normalizing and correcting RNA expression data
A platform to perform normalization and correction on gene expression datasets and combines different datasets into a standard dataset using a framework configured to continuously incorporate new gene expression data. The framework determines a series of conversion factors that are used to on-board new gene expression datasets, such as unpaired datasets, where these conversion factors are able to correct for variations in data type, variations in gene expressions, and variations in collection systems.
1 . A computer-implemented method for measuring gene expression based on RNA expression data from disparate samples, comprising:
obtaining, via one or more processors, (i) an experimental RNA sequence dataset for one or more genes from one or more experimental samples and (ii) a reference RNA sequence dataset for one or more genes from one or more reference samples,
wherein the experimental RNA sequence dataset comprises experimental gene length data, experimental guanine-cytosine (GC) content data, and experimental depth of sequencing data for each of the one or more genes from the one or more experimental samples,
wherein the reference RNA sequence dataset comprises reference gene length data, reference guanine-cytosine (GC) content data, and reference depth of sequencing data for each of the one or more genes from the one or more reference samples,
wherein at least one of the one or more experimental samples and at least one of the one or more reference samples are not from identical tissue type and/or RNA selection method;
generating, via the one or more processors, normalized RNA sequence data to reduce biases introduced by sample preparation by normalizing the experimental depth of sequencing data for the experimental RNA sequence dataset by:
(a) determining, via the one or more processors, a set of ratios of the experimental depth of sequencing data to a geometric mean of the reference depth of sequencing data,
wherein each ratio of the set of ratios represents a respective one of the one or more genes;
(b) determining, via the one or more processors, a size factor as a median of the set of ratios; and
(c) converting, via the one or more processors, the experimental RNA sequence dataset into a normalized RNA sequence dataset using the size factor;
generating, based on the normalized RNA sequence dataset, a gene expression status of the one or more genes of the one or more experimental samples;
generating, via the one or more processors, at least one conversion factor based on a comparison of the normalized RNA sequence dataset to the reference RNA sequence dataset by determining a statistical mapping between sample data of the normalized RNA sequence dataset and sample data of the reference RNA sequence dataset;
integrating, via the one or more processors, the normalized RNA sequence dataset into the reference RNA sequence dataset, by using the at least one conversion factor, to form a uniform gene expression database of comparable gene expression data from the disparate samples; and
enabling a multi-modal machine learning framework to access the uniform gene expression database and integrate the comparable gene expression data with imaging features to generate a neural network output used for analyzing biological samples.
2 . The computer-implemented method of claim 1 , further comprising:
normalizing the experimental gene length data for each gene of the one or more genes from the one or more experimental samples; and
normalizing the experimental GC content data for each gene of the one or more genes from the one or more experimental samples.
3 . The computer-implemented method of claim 2 , wherein normalizing the experimental gene length data comprises using a quantile normalization procedure.
4 . The computer-implemented method of claim 2 , wherein normalizing the experimental GC content data comprises using a quantile normalization procedure.
5 . The computer-implemented method of claim 1 , further comprising:
normalizing, via the one or more processors, the experimental RNA sequence dataset by applying a Reads Per Kilobase Million (RPKM) normalization, a Fragments Per Kilobase Million (FPKM) normalization, or a Transcripts Per Kilobase Million (TPM) normalization.
6 . The computer-implemented method of claim 1 , wherein the experimental RNA sequence dataset is a raw dataset.
7 . The computer-implemented method of claim 1 , wherein the experimental RNA sequence dataset is a Cancer Genome Atlas (TCGA) dataset.
8 . The computer-implemented method of claim 1 , wherein the experimental RNA sequence dataset is a Genotype-Tissue Expression (GTEx) dataset.
9 . The computer-implemented method of claim 1 , wherein determining the statistical mapping comprises determining a linear mapping model between the sample data of the normalized RNA sequence dataset and the sample data of the reference RNA sequence dataset, the computer-implemented method further comprising:
determining an intercept and a beta value for the linear mapping model; and
determining the at least one conversion factor using the statistical mapping from the intercept and the beta value.
10 . The computer-implemented method of claim 1 , wherein generating the at least one conversion factor comprises:
(i) for at least one of the one or more genes, obtaining the sample data from the normalized RNA sequence dataset and obtaining the sample data from the reference RNA sequence dataset;
(ii) determining a linear mapping model between the sample data of the normalized RNA sequence dataset and the sample data of the reference RNA sequence dataset;
(iii) determining an intercept and a beta value for the linear mapping model;
(iv) performing (i)-(iii) a plurality of times for the at least one of the one or more genes to generate a pool of values; and
(v) determining a gene specific conversion factor from a mean intercept and a mean beta value based on the pool of values.
11 . The computer-implemented method of claim 1 ,
wherein an embedding is visualized using a dimensionality reduction technique.
12 . The computer-implemented method of claim 1 , wherein the tissue type is fresh frozen (FF) or formalin fixed, paraffin embedded (FFPE).
13 . The computer-implemented method of claim 1 , wherein the RNA selection method is exon capture or poly-A RNA selection.
14 . A computing device comprising one or more memories and one or more processors configured to:
obtain, via the one or more processors, (i) an experimental RNA sequence dataset for one or more genes from one or more experimental samples and (ii) a reference RNA sequence dataset for one or more genes from one or more reference samples,
wherein the experimental RNA sequence dataset comprises experimental gene length data, experimental guanine-cytosine (GC) content data, and experimental depth of sequencing data for each of the one or more genes from the one or more experimental samples,
wherein the reference RNA sequence dataset comprises reference gene length data, reference guanine-cytosine (GC) content data, and reference depth of sequencing data for each of the one or more genes from the one or more reference samples, and
wherein at least one of the one or more experimental samples and at least one of the one or more reference samples are not from identical tissue type and/or RNA selection method;
generate, via the one or more processors, normalized RNA sequence data to reduce biases introduced by sample preparation by normalizing the experimental depth of sequencing data for the experimental RNA sequence dataset by:
(a) determining, via the one or more processors, a set of ratios of the experimental depth of sequencing data to a geometric mean of the reference depth of sequencing data of the reference RNA sequence dataset,
wherein each ratio of the set of ratios represents a respective one of the one or more genes;
(b) determining, via the one or more processors, a size factor as a median of the set of ratios; and
(c) converting, via the one or more processors, the experimental RNA sequence dataset into normalized RNA sequence dataset using the size factor;
generate, based on the normalized RNA sequence dataset, a gene expression status of the one or more genes of the one or more experimental samples;
generate, via the one or more processors, at least one conversion factor based on a comparison of the normalized RNA sequence dataset to the reference RNA sequence dataset by determining a statistical mapping between sample data of the normalized RNA sequence dataset and sample data of the reference RNA sequence dataset;
integrate, via the one or more processors, the normalized RNA sequence dataset into the reference RNA sequence dataset, by using the at least one conversion factor, to form a uniform gene expression database of comparable gene expression data from the disparate samples; and
enable a multi-modal machine learning framework to access the uniform gene expression database and integrate the comparable gene expression data with imaging features to generate a neural network output used for analyzing biological samples.
15 . The computing device of claim 14 , wherein the one or more processors are configured to generate the normalized RNA sequence data by being configured to:
normalize the experimental gene length data for the one or more genes from the one or more experimental samples; and
normalize the experimental GC content data for the one or more genes from the one or more experimental samples.
16 . The computing device of claim 14 , wherein the one or more processors are configured to determine the statistical mapping by being configured to determine a linear mapping model between the sample data of the normalized RNA sequence dataset and the sample data of the reference RNA sequence dataset, the one or more processors being further configured to:
determine an intercept and a beta value for the linear mapping model; and
determine the at least one conversion factor using the statistical mapping from the intercept and the beta value.