Analytical approaches for the BOAC SNP panel association
QuickTimeTM and a
TIFF (LZW) decompressor
are needed to see this picture.
with progression free survival in myeloma
Brian Van
Van Ness1
Ness Christine Ramos1
Ramos , Majda Haznadar1
Haznadar , John Crowley2
Crowley , Bart Barlogie3
Barlogie , Brian Durie4
Durie , Antje
Hoerring2, Jeff Haessler2, Vipin Kumar1, Michael Steinbach1, David Johnson5, Gareth Morgan5.
1University of Minnesota, 2Cancer Research and Biostatistics, 3University of Arkansas Medical Sciences, 4Aptium Oncology, 5The Royal Cancer Hospital
ABS
AB T
S RAC
A T
C
MET
ME HO
H D
O S
D
·
10,000 random case-control
· Two phase III clinical trials (similar therapies): ECOG E9487 / SWOG/ECOG S9321
permutations center at 50%
The Bank On A Cure (BOAC) has established DNA banks from multiple cooperative and
· SNP associations with progression free survival (PFS): two arms long >3 years( n=73);
accuracy; never above 63%
short <1 year (n=70)
institutional clinical trials, and platforms for examining the association of genetic variations
·
Actual survival comparison: all
(SNPs) with disease risk and outcomes in myeloma. We have previously described the
SNPs reached accuracy of
STATISTICAL APPROACHES
development and content of a novel custom SNP panel that contains 3,404 SNPs in 983
67% in predicting long versus
genes, representing cellular functions and pathways that may influence disease response,
·
Univariate Analysis: single SNP tested against the phenotype for correlaton and ranked.
short survival.
toxicities, complications, and survival. Although survival certainly varies according to tumor
Cannot identify interacting SNPs
·
Subgroup of Non-syn +
heterogeneity (ie. chromosomal abnormalities, gene expression variati
tions) germline
promolign (regulat
)
ory SNP
SNPs
·
Multivariate Analysis: groups of SNPs of size two or more are tested for possible
variations that influence the microenvironment, drug distribution, drug transport and
association with the phenotype. Example: MDR
reached 76% accuracy
metabolism, may also have an association with survival outcomes. To explore SNP
·
P < .0001
Often infeasible in practice due to combinatorial explosion.
associations with progression free survival (PFS) we compared the BOAC SNP profiles of
short term PFS (less than 1 year, n=70) versus long term PFS (greater than 3 years, n=73)
·
Classification methods: use the entire set of SNPs as features to build a model for
differentiating cases from controls.
in two phase III clinical trials (ECOG E9487 and SWOG S9321). A variety of analytical
approaches was undertaken including univariate rank ordering, recursive partitioning, and
Leave One Out Cross
Permutation
support vector machine learning tools (SVM). Each of these approaches has advantages
validation
Testing
Recursive partitioning
and limitations in dealing with type I false positive errors as well as sensitivity and
· Each time a sample is left
· Labels are shuffled for each
· Recursive partitioning performed on top 50 rank ordered SNPs for each trial separately
specificity.
py We included subset validation app
pproaches and randomization of classes to
out and a model is built on
evaluation
· Identify combinations of SNPs that best disting
yguish PFS groups
p
address how robust and predictive different approaches were. From our analysis we
the remaining data
· Each genotype is evaluated on its ability to make a correct prediction, creating a decision
conclude germline genomic variations do have an impact on progression free survival, with
· This process is repeated 10,000
node
a subset of SNPs from the panel reaching 76% predictive association and odds ratios of
· The built model is evaluated
times
· A pruned decision tree is created in which the minimum number of the strongest nodes
survival of 9.6 (CI 4.5, 20.5), p<0.001, using SVM analysis. Based on univariate
on the left out sample
creates a group prediction
· The p-value is determined based on
·The correct classification rate
approaches, we find the most significant variations associated with survival differences
were genes that could be functionally categorized as pharmacologic. The presentation will
· Best suited when the
the number of times the permutation
focus on the analytical approaches, and refinements necessary to assure predictive value
number of samples are
accuracy has crossed the accuracy
Rs174680 Catechol-O-methyltransferase
compared to random associations. Notwithstanding the clear importance of tumor cell
small
obtained over actual labels
Rs35683 Ghrelin
variations in genetic deregulation, we conclude that various functions within the bone
Rs1047643 Farnesyl transferase 1
marrow and drug response likely interplay
interplay as a com
complex
plex influence on
on disease
disease progression
progression,
RESULT
RESUL S
Rs215101 ATP binding cassette C (ABCC)
response, and survival.
Univariate Permutation Test P-Value
CHIP DESIGN
·
Individual SNP associations
with true phenotype are not
· Bank On a Cure (BOAC) Mission: to create a DNA bank and develop genetic
distinguishable from random
CONCLUSIONS
correlates with myeloma risk, progression, response and toxicities associated
permutation of phenotype
with therapies.
·
Univariate analysis shows associations for individual SNPs within random
·
A combination of SNPs may
permutations
be mo
more
re predictive
predictiv than
than
Fig 1. An approach taken in designing
Table1. Functional categories on the
individual SNPs
·
Classification methods that look at all SNPs or subgroups show association with
the custom SNP chip
SNP chip panel
survival with accuracy significantly above random
Functional Category
#Genes #SNPs
ADME/DMET
130
455
·
Recursive partitioning shows some weakly associated individual SNPs
Cancer
406
1558
Carbohydrate Metabolism
69
384
·
Interactions of SNPs suggest multiple germline gene variations may contribute to
Cell Cycle
230
867
Cell Death
433
1662
Performance for Selected Categories of SNPs
survival
Cell Signaling
90
352
Cell-To-Cell Signaling and Interaction
248
880
Non
Introns
Synonymous
Admixture
UTR
Other
Accuracy
Cellular Growth and Proliferation
420
1451
·
Patient outcomes are likely affected by tumor variations and germline variations
synonymous
Intergenic
(%)
Cellular Movement
227
923
DNA Replication, Recombination, and Repair
204
854
66.43
Drug
Drug Me
Metabol
tabo ism
lism
20
114
11
Gene Expression
240
951
58.74
ACKNOWLEDGMENTS
Hematological Disease
223
876
Immune Response
247
985
51.74
Lipid Metabolism
146
664
· Van Ness, B. et al. Genomic variation in myeloma: design, content, and initial application of the Bank On A Cure SNP Panel to
Molecular Transport
170
708
72.72
detect associations with progression-free survival. BMC Med 6, 1-14 (2008).
Nucleic Acid Metabolism
30
161
Skeletal and Muscular Disorders
64
289
· Support from the International Myeloma Foundation and NCI P01 51112
71.33
Skeletal and Muscular System Development and Function
77
278
Signaling Kinase, Phosphatase, Transferase
198
885
54.54
Inflammation & Immunity
196
813
No conflicts of interest to report for any author.
3404 SNPs in total of 983 genes
69.99