Upcycling genomics data:

From publicly available "junk" to priceless "treasure"

Shannon E. Ellis

​

What makes primary cancer different than metastatic cancer?

www.hopkinsmedicine.org

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

3~6 months

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

3~6 months

1-2 wks

2-4 wks

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

33-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Analyze data and answer biological question

Data cleaning

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

33-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Total: 2+ years

Analyze data and answer biological question

Data cleaning

Biologists have recently gotten pretty good at making their data available to the public.

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

Data cleaning

33-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Total: 1+ years

Analyze data and answer biological question

...but they’re not great at making these data easily accessible and well-annotated.

Biologists have recently gotten pretty good at making their data available to the public.

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

Data cleaning

33-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Total: 1+ years

Analyze data and answer biological question

Project Goal: Take publicly available RNA-Seq data and make it available and easy-to-use

Easy to Get

Comprehensive

Useful for future study

Genetics101

The Central Dogma of Genetics

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from jeff leek

TGACTGGATCTAGTCAGCTAGCTAGCATATGCTAATGTTTTAGTAGCCGTA

The Central Dogma of Genetics

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from jeff leek

The Central Dogma of Genetics

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from alyssa frazee

slide adapted from jeff leek

gene

The Central Dogma of Genetics

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from alyssa frazee

slide adapted from jeff leek

gene

exons

The Central Dogma of Genetics

AUCAGUCGAUCACCGAU

transcription

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from alyssa frazee

slide adapted from jeff leek

RNA

The Central Dogma of Genetics

AUCAGUCGAUCACCGAU

transcription

translation

ACTGACCTAGATCAGTCGATCGATCGTATACGATTACAAAATCATCGGCAT

DNA

slide adapted from alyssa frazee

slide adapted from jeff leek

RNA

proteins

Two copies of DNA -> many transcripts -> many proteins

blueprint

role

in the cell

# copies/cell

2

RNA

proteins

DNA

functional

unit

gene

# unique functional

units

20,000

Two copies of DNA -> many transcripts -> many proteins

blueprint

carry out cellular functions

2

varies

~10¹⁰

RNA

proteins

DNA

gene

proteins

(metabolites, hormones, etc.)

20,000

~100,000

role

in the cell

# copies/cell

functional

unit

# unique functional

units

Two copies of DNA -> many transcripts -> many proteins

blueprint

messenger

carry out cellular functions

2

varies

~360,000

varies

~10¹⁰

RNA

proteins

DNA

gene

transcript

proteins

(metabolites, hormones, etc.)

20,000

~100,000

~100,000

role

in the cell

# copies/cell

functional

unit

# unique functional

units

Variability at the level of RNA allows for a heart cell to function differently than a brain cell

!=

Measuring � RNA levels

slide adapted from jeff leek

Next Generation Sequencing (NGS) Has Completely Revolutionized How We Study Genetics

Next Generation Sequencing (NGS) in one slide

RNA

Step 1: Extract RNA to get sample of interest

Next Generation Sequencing (NGS) in one slide

RNA

Step 1: Extract RNA to get sample of interest

Step 2: Chop up RNA into smaller pieces

Next Generation Sequencing (NGS) in one slide

RNA

Step 1: Extract RNA to get sample of interest

Step 3: Sequence the sample

Step 2: Chop up RNA into smaller pieces

Next Generation Sequencing (NGS) in one slide

RNA

Step 1: Extract RNA to get sample of interest

Step 3: Sequence the sample

Step 3: Obtain short read data from the sequencer

Step 2: Chop up RNA into smaller pieces

Next Generation Sequencing (NGS) in one slide

RNA

AUCAGUCGAUCACCGAU

A short read tells you the sequence of the RNA in that read

slide adapted from jeff leek

slide adapted from jeff leek

Sequence Identifier

Sequence

Quality scores

We’ve got 40M+ reads.

What does that all mean?

We first need to align these reads back to the genome

Genome

(DNA)

slide adapted from jeff leek

The Human Genome Project (HGP) determined the reference sequence of the human genome in 2001.

We first need to align these reads back to the genome

Genome

(DNA)

slide adapted from jeff leek

Nellore et al. (2016)

Bioinformatics

http://rail.bio/

​

We can then align each short read back to the reference genome to figure out where it came from.

We first need to align these reads back to the genome

coverage vector

2

6

0

11

6

Genome

(DNA)

slide adapted from jeff leek

The number of reads at each position lets us know abundance

We first need to align these reads back to the genome

coverage vector

2

6

0

11

6

Genome

(DNA)

slide adapted from jeff leek

We then summarize across all the positions in a gene to get an estimate for gene expression.

RNA-Seq = estimate expression across entire genome

estimate expression!

RNA-Seq = estimate expression across entire genome

estimate expression!

expression ≅ # RNA-Seq reads

Project Goal: Take publicly available RNA-Seq data and make it available and easy-to-use

Easy to Get

Comprehensive

Useful for future study

GTEx

https://commonfund.nih.gov/GTEx

TCGA

GTEx

https://commonfund.nih.gov/GTEx

SRA

We’ll take these ~70,000 samples, align each back to the reference genome, and then, for each sample, we’ll estimate expression across the genome.

Project Goal: Take publicly available RNA-Seq data and make it available and easy-to-use

Easy to Get

Comprehensive

Useful for future study

X

�expression data for ~70,000 human samples

samples

expression estimates

�expression data for ~70,000 human samples

samples

expression estimates

gene

exon

junctions

ERs

Find a researcher with access to patient samples

What makes primary cancer different than metastatic cancer?

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

Data cleaning

33-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Determine expression differences between primary cancer and metastatic cancer

Find publicly available RNA-sequencing data from primary cancer and metastasis

Total: 1+ years

Find a researcher with access to patient samples

Collect patient samples and information

Extract DNA/RNA from samples

Sequence samples

Process sequencing data

Data cleaning

3-6 months

1-3 months

1 month - 1+ years

3~6 months

1-2 wks

2-4 wks

Determine expression differences between primary cancer and metastatic cancer

Get already processed and summarized RNA-Seq data from recount2

Total: months

What makes primary cancer different than metastatic cancer?

Since September of this year, 555 different people have accessed data in recount2

These 555 people have accessed 417,488 files (91,003 unique) in recount2

Project Goal: Take publicly available RNA-Seq data and make it available and easy-to-use

Easy to Get

Comprehensive

Useful for future study

X

X

�expression data for ~70,000 human samples

Answer meaningful questions about human biology and expression

samples

expression estimates

gene

exon

junctions

ERs

GTEx

N=9,962

TCGA

N=11,284

SRA

N=49,848

�expression data for ~70,000 human samples

samples

phenotypes

?

samples

expression estimates

gene

exon

junctions

ERs

Answer meaningful questions about human biology and expression

GTEx

N=9,962

TCGA

N=11,284

SRA

N=49,848

SRA phenotype information is far from complete

SRA

SRA phenotype information is far from complete

z

z

z

SRA

Even when information is provided, it’s not always clear…

SRA

Even when information is provided, it’s not always clear…

“1 Male, 2 Female”, “2 Male, 1 Female”, “3 Female”, “DK”, “male and female” “Male (note: ….)”, “missing”, “mixed”, “mixture”, “N/A”, “Not available”, “not applicable”, “not collected”, “not determined”, “pooled male and female”, “U”, “unknown”, “Unknown”

SRA

Even when information is provided, it’s not always clear…

“1 Male, 2 Female”, “2 Male, 1 Female”, “3 Female”, “DK”, “male and female” “Male (note: ….)”, “missing”, “mixed”, “mixture”, “N/A”, “Not available”, “not applicable”, “not collected”, “not determined”, “pooled male and female”, “U”, “unknown”, “Unknown”

SRA

in-silico Phenotyping

slide adapted from jeff leek

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Machine Learning: Making predictions from data

Data Set #1

Data Set #2

Machine Learning: Making predictions from data

Data Set #1

Data Set #2

Training Data

Validation

Data

Machine Learning: Making predictions from data

Data Set #1

Data Set #2

Training Data

Data used to build the predictor

We’re interested in predicting phenotype from expression data...

There are a number of different curves you could fit through these data

This curve fits every point in the training data perfectly...

Machine Learning: Making predictions from data

Data Set #1

Data Set #2

Training Data

Data used to build the predictor

Validation

Data

Samples held back from training

What if we tried to predict phenotype in the validation data?

The curve no longer fits every point perfectly.

The curve no longer fits every point perfectly.

Machine Learning: Making predictions from data

Data Set #1

Data Set #2

Training Data

Data used to build the predictor

Validation

Data

Test

Data

Samples held back from training

Independent data set to test predictor

We can now test prediction accuracy in an independent set of samples.

The line generated from the training data accurately predicts phenotype in the test data

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

build and optimize phenotype predictor

Training Data

Missingness limited in GTEx phenotype data

GTEx

Missingness limited in GTEx phenotype data

GTEx

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

build and optimize phenotype predictor

Training Data

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

divide samples

build and optimize phenotype predictor

test accuracy of predictor

Training Data

Validation

Data

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

predict phenotypes across samples in TCGA

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

divide samples

build and optimize phenotype predictor

test accuracy of predictor

Training Data

Validation

Data

Test

Data

Goal :��to accurately predict critical phenotype information for all samples in recount2

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

divide samples

build and optimize phenotype predictor

predict phenotypes across SRA samples

test accuracy of predictor

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

predict phenotypes across samples in TCGA

Training Data

Validation

Data

Test

Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

Prediction is done using linear regression

New sample’s expression

Prediction is done using linear regression

New sample’s expression

Prediction is done using linear regression

New sample’s expression

New sample’s predicted phenotype

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

92.7%

filter_regions()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

phenopredict

92.7%

filter_regions()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

phenopredict

Set of discriminatory regions

92.7%

filter_regions()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

phenopredict

Set of discriminatory regions

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

92.7%

filter_regions()

build_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

phenopredict

Set of discriminatory regions

Relationship between expression and phenotype

92.7%

filter_regions()

build_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

phenopredict

test_predictor()

predictions

Predict phenotype and assess accuracy in training set data

Relationship between expression and phenotype

Set of discriminatory regions

Likelihood of phenotype for each individual

92.7%

filter_regions()

build_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

phenopredict

test_predictor()

predictions

Predict phenotype and assess accuracy in training set data

Relationship between expression and phenotype

Set of discriminatory regions

Assign phenotype to most likely category

92.7%

filter_regions()

build_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

phenopredict

test_predictor()

predictions

Predict phenotype and assess accuracy in training set data

Relationship between expression and phenotype

Set of discriminatory regions

Assign phenotype to most likely category

92.7%

filter_regions()

build_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

phenopredict

test_predictor()

predictions

Predict phenotype and assess accuracy in training set data

Relationship between expression and phenotype

Set of discriminatory regions

Assign phenotype to most likely category

92.7%

filter_regions()

test_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

predictions

reported

phenopredict

Predict phenotype and assess accuracy in training set data

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

92.7%

filter_regions()

test_predictor()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

predictions

reported

Prediction accuracy: 100%

phenopredict

Predict phenotype and assess accuracy in training set data

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

filter_regions()

test_predictor()

extract_data()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

Identify regions with differential expression for each level

predictions

reported

Extract expression information at regions identified by filter_regions() in a new data set

Prediction accuracy: 100%

expression @ filtered regions

new data set samples

92.7%

sex

expression

male

female

phenopredict

Predict phenotype and assess accuracy in training set data

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

92.7%

filter_regions()

test_predictor()

extract_data()

predict_pheno()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

predictions

reported

Extract expression information at regions identified by filter_regions() in a new data set

Predict phenotypes across samples in this new data set

Prediction accuracy: 100%

expression @ filtered regions

new data set samples

apply coefficient estimates to the extracted data

phenopredict

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

Predict phenotype and assess accuracy in training set data

92.7%

filter_regions()

test_predictor()

extract_data()

predict_pheno()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

predictions

reported

Extract expression information at regions identified by filter_regions() in a new data set

Predict phenotypes across samples in this new data set

Prediction accuracy: 100%

expression @ filtered regions

new data set samples

apply coefficient estimates to the extracted data

phenopredict

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

Predict phenotype and assess accuracy in training set data

92.7%

filter_regions()

test_predictor()

extract_data()

predict_pheno()

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

sex

expression

male

female

Identify regions with differential expression for each level

predictions

reported

Extract expression information at regions identified by filter_regions() in a new data set

Predict phenotypes across samples in this new data set

Prediction accuracy: 100%

expression @ filtered regions

new data set samples

apply coefficient estimates to the extracted data

predictions in new data set

phenopredict

build_predictor()

Extract coefficient estimates across regions

expression ~ phenotype

filtered regions (r)

male

female

phenotype (P)

Predict phenotype and assess accuracy in training set data

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

Accuracy

100%

100%

100%

100%

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

Validation

Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

100%

100%

100%

100%

Accuracy

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

Validation

Data

Test

Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

100%

100%

100%

100%

Accuracy

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

Validation

Data

Test

Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

Make predictions!

100%

100%

100%

100%

Accuracy

​

gene, exon, exon-exon junction and expressed region RNA-Seq data

SRA

Sequence Read Archive

N=49,848

​

TCGA

The Cancer Genome Atlas

N=11,284

GTEx

Genotype Tissue Expression Project

N=9,662

Training Data

Validation

Data

Test

Data

filter_regions()

build_predictor()

test_predictor()

extract_data()

predict_pheno()

functions

phenopredict

Make predictions!

100%

100%

100%

100%

Accuracy

99.9%

Sex prediction is accurate across data sets

99.8%

99.0%

86.3%

99.9%

Sex prediction is accurate across data sets

99.8%

99.0%

86.3%