A key focus in current cancer research is the discovery of cancer biomarkers that allow earlier detection with high accuracy and lower costs for both patients and hospitals. Blood samples have long been used as a health status indicator, but DNA methylation signatures in blood have not been fully appreciated in cancer research. Historically, analysis of cancer has been conducted directly with the patient’s tumor or related tissues. Such analyses allow physicians to diagnose a patient’s health and cancer status; however, physicians must observe certain symptoms that prompt them to use biopsies or imaging to verify the diagnosis. This is a post-hoc approach. Our study will focus on epigenetic information for cancer detection, specifically information about DNA methylation in human peripheral blood samples in cancer discordant monozygotic twin-pairs. This information might be able to help us detect cancer much earlier, before the first symptom appears. Several other types of epigenetic data can also be used, but here we demonstrate the potential of blood DNA methylation data as a biomarker for pan-cancer using SAS® 9.3 and SAS® EM. We report that 55 methylation CpG sites measurable in blood samples can be used as biomarkers for early cancer detection and classification.

1. Pan-Cancer Epigenetic Biomarker Selection
from Blood Samples Using SAS®
X I C H E N
D E PA RT M E N T O F M O L EC U L A R A N D C E L LU L A R B I O LO GY
U N I V E RS I T Y O F K E N T U C KY
J I N X I E
D E PA RT M E N T O F STAT I ST I C S
U N I V E RS I T Y O F K E N T U C KY
Q I N G CO N G Y UA N
A S S I STA N T P RO F ES S O R
D E PA RT M E N T O F STAT I ST I C S
M I A M I U N I V E RS I T Y
10/2/2018 #MWSUG2018 ##HS45

2. 10/2/2018 #MWSUG2018 ##SSXX 2
Graduate bioinformatician/biochemistian with
Statistic MA certification completing the final year of
a Ph.D degree. Passionate about Big Data, Machine
Learning and AI research, with strong technical and
interpersonal skills, adept at working in teams and
successfully delivering projects. I was awarded
Microsoft Azure Research Funding in 2016 and
gained a Grow with Google Challenge
Scholarship in 2018.I am a certified SAS
programmer and certified Deep Learning and Cuda
instructor.

3. 10/2/2018 #MWSUG2018 ##SSXX 3
Machine
Learning and
Deep
Learning

4. Leading Causes
of Death in US
10/2/2018 #MWSUG2018 ##HS45 4
In 2014, CDC reported Cancer is
the second leading cause of US
Death (heart disease is #1)
Cancers can be caused by
genetics.
However, genetic defect doesn’t
mean cancer.
Twin-study shows one identical
twin could have cancer and the
other cancer-free.

5. The new science of epigenetics reveals how your
choices you make can change your genes—and
those of your kids.10/2/2018 #MWSUG2018 ##HS45 5

6. 10/2/2018 #MWSUG2018 ##HS45 6
Epigenetics were long-
known from at least
1970s, but until 90’s,
epigenetic phenomena
were regarded as a
sideshow to the main
event, DNA.
“Biochemical switches”
for genes that play a role
in many diseases—
including cancer,
schizophrenia, autism,
Alzheimer's diabetes and
many others—to lie
dormant.
In this presentation, we
will focus only on one
type of epigenetic
mechanism—DNA
methylation
Epigenetics in Action

7. Epigenetics: DNA Methylation
10/2/2018 #MWSUG2018 ##HS45 7

8. 10/2/2018 #MWSUG2018 ##HS45 8
DNA Methylation and Cancer

9. 10/2/2018 #MWSUG2018 ##HS45 9
DNA
Methylation
and Cancer

10. Cancer Diagnosis
10/2/2018 #MWSUG2018 ##HS45 10
Traditional diagnosis pipeline
Early Detection
via
Blood Sample
Future diagnosis pipeline

11. Data Source
10/2/2018 #MWSUG2018 ##HS45 11
GEO dataset link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE89093.

12. Cancer typing distribution
10/2/2018 #MWSUG2018 ##HS45 12

13. Data
•The feature dimension of the data was rather large: 92 observations
(rows) by 453,627 features (columns)
•To validate the data, we generate statistics:
(1) moments,
(2) basic measures of location and variability,
(3) tests for location,
(4) quantiles,
(5) extreme observations.
Aim: feature selection
10/2/2018 #MWSUG2018 ##HS45 13

14. 10/2/2018 #MWSUG2018 ##HS45 14
Variable
Screen step
A. Sure Independence
Screening;
B. Independence Screening
via Distance Correlation;
C. T-test Analysis;
D. Expectation of the
Conditional Difference.
Red line in the figure shows
the first 100 variables
according to the ranking of
each approach.

15. Proposed Epigenetic Biomarkers
10/2/2018 #MWSUG2018 ##HS45 15
•Identified 55 common CpG sites among the top 100 candidates identified by each of the four
different methods described above.
•Map the epigenetic CpG sites back to the genetic data.
CpG ID
Associated
Gene
Gene Important/Function
cg23412136 MIR548N
Non-coding region (microRNAs), involved in
post-transcriptional regulation of gene
expression.
cg09720012 PAPSS1
Diseases associated with PAPSS1 include
Chronic Monocytic Leukemia.
cg17964016 DEPTOR Inhibits the p53 kinase activity.
cg16824024 PPP4R2
Related to pathways of DNA double-strand
break repair.

16. Model performance (evaluation)
summary
Data Set Accuracy FPR FNR
Training
Set (70%)
100% 0% 0%
Testing
Set (30%)
71.4% 42.9% 14.3%
10/2/2018 #MWSUG2018 ##HS45 16
Receiver operating characteristic (ROC) curve generated
using linear discriminate analysis (LDA) for the testing
dataset examining the 55 common CpG sites selected
from the variable screening step.

17. Future Direction: Deep
Learning Approach
10/2/2018 #MWSUG2018 ##SSXX 17

18. Future Direction: Deep
Learning Approach
10/2/2018 #MWSUG2018 ##SSXX 18
14,157
60,483
Data Processing

19. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 19Image source: https://themenwhostareatcodes.files.wordpress.com/2014/03/
Neuron

20. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 20

21. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 21

22. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 22

23. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 23Image source: https://themenwhostareatcodes.files.wordpress.com/2014/03/

24. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 24

25. Future Direction: Deep Learning
Approach
10/2/2018 #MWSUG2018 ##HS45 25

26. Preliminary Results
10/2/2018 #MWSUG2018 ##HS45 26
TrueLabels Predicted Labels
Cancer detection results from Lung Cancer.
Cancer classification results
for every cancer types.

27. 10/2/2018 #MWSUG2018 ##HS45

28. Name: Xi (Bill) Chen
Organization: University of Kentucky
Work Phone: 8572421588
E-Mail: billchenxi@gmail.com
Web: www.uky.edu
Twitter: @billchenxi
10/2/2018 #MWSUG2018 ##HS45 28