津田班 tsuda-crest 2015/12/24 京大 山田 yamada, kyoto-u
DESCRIPTION
LAMPLINK; a toy case Candidate gene data set in DM 2000 case 2000 controls 37 SNPs RUN LAMPLINK with data commmand./lamplink --file./cg/cg --lamp --model-dom --sglev upper out./cg/cgdom -- fisher options model dom: minor allele is dominant; sglev (value): set statistical significance level (default 0.05) upper(value): set MAF value 0.5 fisher: use Fisher’s exact test LAMPLINK result in lamp fileTRANSCRIPT
津田班Tsuda-CREST
2015/12/24京大 山田Yamada, KYOTO-U
LAMP(1)LAMPLINK
• on RA GWAS • LAMPLINK with parallel implementation
– On going…
LAMPLINK; a toy caseCandidate gene data set in DM• 2000 case 2000 controls • 37 SNPsRUN LAMPLINK with datacommmand• ./lamplink --file ./cg/cg --lamp --model-dom --sglev 0.05 --upper 0.5 --out ./cg/cgdom --
fisheroptionsmodel dom: minor allele is dominant;sglev (value): set statistical significance level (default 0.05)upper(value): set MAF value 0.5fisher: use Fisher’s exact test LAMPLINK result in lamp file
LAMPLINKCandidate gene data set in DM• 2000 case 2000 controls • 37 SNPsRUN LAMPLINK with datacommmand• ./lamplink --file ./cg/cg --lamp --model-dom --sglev 0.05 --upper 0.5 --out ./cg/cgdom --
fisheroptionsmodel dom: minor allele is dominant;sglev (value): set statistical significance level (default 0.05)upper(value): set MAF value 0.5fisher: use Fisher’s exact test LAMPLINK result in lamp file
LAMPLINK; p-distribution
Many SNPs in an Intermediate LD locus: SNP-pair p-val dist
Small p-dominant
Large p-dominant
Mean of 100 experimentsof SNP-pair p-vals
Mean: 0.498
LAMP(2)
• P-value distribution of n-arities for Data sets with LD/dependencies– Uniform– Large-p dominant– Small-p dominant
• Locus-wise, Case-wise p-dist heterogeneity– If repeated, overall p-dist seems uniform
• Q: Anything should be done for the interpretation of nominal p in this deviated distribution or not?
LAMPLINKCandidate gene data set in DM• 2000 case 2000 controls • 37 SNPsRUN LAMPLINK with datacommmand• ./lamplink --file ./cg/cg --lamp --model-dom --sglev 0.05 --upper 0.5 --out ./cg/cgdom --
fisheroptionsmodel dom: minor allele is dominant;sglev (value): set statistical significance level (default 0.05)upper(value): set MAF value 0.5fisher: use Fisher’s exact test LAMPLINK result in lamp file
Combinations ~ Itemsets ~ZDD(1)
• Nysol; ZDD– ZDD objects of 2x2 tables’ marginal counts and [1,1] cell
counts– ZDD object algebra (+ - * / % > < == !=)– Statistical test calculations are too complicated to be
done with ZDD object algebra but effect size calculations can be done.
–Q: How to bring the results to “statistical judgement”? : Monte-carlo permuation?
• Under the trial
Combinations ~ Itemsets ~ZDD(2)
• For multiple phenotypes, combination analyses results can be stocked in the shape of ZDD.
• Inter-phenotype results on combinations might be integrated with ZDD.
Combinations, Simplices
• Anything good to use simplex symmetricity ?
Simplex
Test vectors and “LAMPLINK combination tests” in 2x4 table
Simplex’s geometric features (symmetricity)
might be utilized … how???• LAR (Least Angle Regression)
– Returns similar results with Lasso– http://
d.hatena.ne.jp/isseing333/20110309/1299675311
LAR on simplex
• Because LAR uses “angles” and simplex-representation has obvious angle information in it, LAR might be solved in quicker/lighter way???
• In case of “discrete” state, LAR’s output should be “intergers” or “rationals” with “AND/OR” logic.
****
*
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
|beta|/max|beta|
Sta
ndar
dize
d C
oeffi
cien
ts
****
*
****
*
****
*
LAR
17
****
*
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
|beta|/max|beta|
Sta
ndar
dize
d C
oeffi
cien
ts
****
*
****
*
****
*
LASSO
17