Post on 10-Jun-2020
Statistics for high-dimensional data:P-values and Stability Selection
Peter Buhlmann and Sara van de Geer
Seminar fur Statistik, ETH Zurich
May 2012
P-values for high-dimensional linear models
Y = Xβ0 + ε
goal: statistical hypothesis testing
H0,j : β0j = 0 or H0,G : β0
j = 0 for all j ∈ G
background: if we could handle the asymptotic distribution ofthe Lasso β(λ) under the null-hypothesis; could construct p-values
this is very difficult!asymptotic distribution of β has some point mass at zero,...Knight and Fu (2000) for p <∞ and n→∞;Montanari (2010, 2012) ... for random design with i.i.d. columnsnot practical at the moment
Variable selection
Example: Motif regressionfor finding HIF1α transcription factor binding sites in DNA seq.Muller, Meier, PB & Ricci
Yi ∈ R: univariate response measuring binding intensity ofHIF1α on coarse DNA segment i (from CHIP-chip experiments)Xi = (X (1)
i , . . . ,X (p)i ) ∈ Rp:
X (j)i = abundance score of candidate motif j in DNA segment i
(using sequence data and computational biology algorithms,e.g. MDSCAN)
question: relation between the binding intensity Y and theabundance of short candidate motifs?
; linear model is often reasonable“motif regression” (Conlon, X.S. Liu, Lieb & J.S. Liu, 2003)
Y = Xβ + ε, n = 287, p = 195
goal: variable selection; find the relevant motifs among the p = 195 candidates
Motif regression
for finding HIF1α transcription factor binding sites in DNA seq.
Yi ∈ R: univariate response measuring binding intensity oncoarse DNA segment i (from CHIP-chip experiments)X (j)
i = abundance score of candidate motif j in DNA segment i
variable selection in linear model Yi = β0 +
p∑j=1
βjX(j)i + εi ,
i = 1, . . . ,n = 287, p = 195
; Lasso selects 26 covariates and R2 ≈ 50%i.e. 26 interesting candidate motifs
motif regression: estimated coefficients β(λCV)
0 50 100 150 200
0.00
0.05
0.10
0.15
0.20
original data
variables
coef
ficie
nts
which variables in S are false positives?(p-values would be very useful!)
(Multi) Sample splittingwhich was an early but sub-ideal proposal
I select variables on first half of the sample ; SI compute OLS for variables in S on second half of the
sample; P-values Pj based on Gaussian linear model
if j ∈ S : Pj from t-statistics
if j /∈ S : Pj = 1 (i.e. if β(j) = 0)
Bonferroni-“style” corrected P-values:
Pcorr,j = min(Pj · |S|,1)
; (conserv.) familywise error control withPcorr,j (j = 1, . . . ,p)
(Wasserman & Roeder, 2008)
this is a “P-value lottery”motif regression example: p = 195, n = 287
adjusted P-values for same important variableover different random sample-splits
ADJUSTED P−VALUE
FR
EQ
UE
NC
Y
0.0 0.2 0.4 0.6 0.8 1.0
020
4060
80
; improve by aggregating over many sample-splits(which improves “efficiency” as well)
this is a “P-value lottery”motif regression example: p = 195, n = 287
adjusted P-values for same important variableover different random sample-splits
ADJUSTED P−VALUE
FR
EQ
UE
NC
Y
0.0 0.2 0.4 0.6 0.8 1.0
020
4060
80
; improve by aggregating over many sample-splits(which improves “efficiency” as well)
Sample splitting multiple times
run the sample-splitting procedure B times and do a non-trivialaggregation of p-values
P-values: P(1)j , . . . ,P(B)
j
goal:aggregation of P(1)
j , . . . ,P(B)j to a single P-value Pfinal,j
problem: dependence among P(1)j , . . . ,P(B)
j
define
Q(j)(γ) = qγ︸︷︷︸emp. γ-quantile fct.
(P(j)corr,b/γ; b = 1, . . .B)
e.g: γ = 1/2, aggregation with the median; (conserv.) familywise error control for any fixed value of γ
what is the best γ? it really matters; can “search” for it an correct with an additional factor
“adaptively” aggregated P-value:
P(j)final = (1− log(γmin)) · inf
γ∈(γmin,1)Q(j)(γ)
Q(j)(γ) = qγ(P(j)corr,b/γ; b = 1, . . .B)
; reject H(j)0 : βj = 0 ⇐⇒ P(j)
final ≤ α
P(j)final equals roughly a raw P-value based on sample size bn/2c,
multiplied by
a factor ≈ (5− 10) · |S|(which is to be compared with p)
for familywise error rate (FWER) =P[at least one false positive selection]
TheoremConsider Gaussian linear model (with fixed design) andassume:
I limn→∞ P[S ⊇ S0] = 1 screening propertyI |S| < bn/2c sparsity property
Then:strong control for either familywise error rate or for falsediscovery rate
motif regression examplep = 195, n = 287
●●●●●●●●●●●●●
●
●●●●●●●●
●
●●
●
●
●
●●●●●●●●●●●●●●●●●●●
●
●●●●
●
●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●
●●
●
●
●●●●●●●●●●
●
●●●●●●●●
●
●●●
●
●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
0 50 100 150 200
0.00
0.05
0.10
0.15
0.20
motif regression
variables
coef
ficie
nts
●
●
◦: variable/motif with FWER-adjusted p-value 0.006◦: p-value clearly larger than 0.05
(this variable corresponds to known true motif)
discussion: multi sample splitting
I assumes P[S ⊇ S0]→ 1; requires the beta-min condition
such an assumption should be avoided for hypothesistesting(because this is the essence of the question in testingwhether β0
j is smallish or sufficiently large)I necessarily requires design conditions; but this is
unavoidable
Stability Selection (Meinshausen & PB, 2010)which allows to go way beyond linear models
selection of “features” from the set {1, . . . ,p}, e.g.:I variable selection in regression or classificationI edge selection in a graphI membership to a clusterI ...
selection procedure
Sλ ⊆ {1, . . . ,p},λ a tuning parameter
prime example: Lasso for selecting variables in a linear model
subsampling:I draw sub-sample of size bn/2c without replacement,
denoted by I∗ ⊆ {1, . . . ,n}, |I∗| = bn/2cI run the selection algorithm Sλ(I∗) on I∗
I do these steps many times and compute therelative selection frequencies
Πλj = P∗(j ∈ Sλ(I∗)), j = 1, . . . ,p
(P∗ is w.r.t. sub-sampling)
could also use bootstrap sampling with replacement...
subsampling:I draw sub-sample of size bn/2c without replacement,
denoted by I∗ ⊆ {1, . . . ,n}, |I∗| = bn/2cI run the selection algorithm Sλ(I∗) on I∗
I do these steps many times and compute therelative selection frequencies
Πλj = P∗(j ∈ Sλ(I∗)), j = 1, . . . ,p
(P∗ is w.r.t. sub-sampling)
could also use bootstrap sampling with replacement...
if we consider many regularization parameters:
{Sλ; λ ∈ Λ}
(Λ can be discrete or continuous)
Sstable = {j ; maxλ∈ΛΠλj ≥ πthr}
see also Bach (2009) for a related proposal
How to choose the threshold πthr?
denote by V = |Sc0 ∩ Sstable| = number of false positives
consider a selection procedure which selects q variables(e.g. top 50 variables when running Lasso over many λ’s)
Theorem (Meinshausen & PB, 2010main assumption: exchangeability conditionin addition: S is better than “random guessing”Then:
E[V ] ≤ 12πthr − 1
q2
p
i.e. finite sample control, even if p � n; choose threshold πthr to control e.g. E[V ] ≤ 1 orP[V > 0] ≤ E[V ] ≤ α
note the generality of the Theorem...
I it works for any method which is better than “randomguessing”
I it works not only for regression but also for “any” discretestructure estimation problem (whenever there is ainclude/exclude decision); variable selection, graphical modeling, clustering, ...
and hence there must be a fairly strong condition...Exchangeability condition:the distribution of {I{j∈Sλ}; j ∈ SC
0 } is exchangeablenote: only some requirement for noise variables
note the generality of the Theorem...
I it works for any method which is better than “randomguessing”
I it works not only for regression but also for “any” discretestructure estimation problem (whenever there is ainclude/exclude decision); variable selection, graphical modeling, clustering, ...
and hence there must be a fairly strong condition...Exchangeability condition:the distribution of {I{j∈Sλ}; j ∈ SC
0 } is exchangeablenote: only some requirement for noise variables
Many simulations!
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
data p n
snr s
violations max cor
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
0
1/4
1/2
3/4
1
P(f
irst 0
.1s
corr
ect)
X
0.5
4100
1000(A)
0
0.34
X
2
X
0.5
10
2
X
2
X
0.5
201000
0
0.11
X
2
X
0.5
50
0
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
4200
1000(B)
0
0.6
X
2
X
0.5
10
5
X
2
X
0.5
201000
1
0.54
X
2
X
0.5
50
44
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
4200
1000(C)
14
0.99
X
2
X
0.5
10
108
X
2
X
0.5
201000
131
0.99
X
2
X
0.5
50
302
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
10200100(D)
1
0.7
X
2
X
0.5
30
4
X
2
X
0.5
4200
1000
2
0.69
X
2
X
0.5
10
11
X
2
X
0.5
201000
0
0.69
X
2
X
0.5
50
2
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
10200200(E)
66
0.73
X
2
X
0.5
30
37
X
2
X
0.5
4200
1000
103
0.78
X
2
X
0.5
10
342
X
2
X
0.5
201000
259
0.77
X
2
X
0.5
50
62
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
42587660(F)
0
0.48
X
2
X
0.5
10
4
X
2
X
0.5
20
70
X
2
X
0.5
50
159
X
2
rep(1, 2)
rep(
1, 2
)
rep(1, 2)
rep(
1, 2
)
X
0.5
4115
4088(G)
118
0.82
X
2
X
0.5
10
477
X
2
rep(
1, 2
)
rep(
1, 2
)
rep(
1, 2
)
rep(
1, 2
)
0
1/4
1/2
3/4
1
P(f
irst 0
.4s
corr
ect)
X
X
X
X
X X
X
X
rep(
1, 2
)
rep(
1, 2
)
X
X
X
X
X
X
X X
rep(
1, 2
)
rep(
1, 2
)
X
X
X X X X X X
rep(
1, 2
)
rep(
1, 2
)
X
X
XX
X
X
X
X
X
X
X
X
rep(
1, 2
)
rep(
1, 2
)
X
X
X X X
X
XX
XX
X X
rep(
1, 2
)
rep(
1, 2
)
X X X X
X
X
X X
rep(
1, 2
)
rep(
1, 2
)
X
X
X X
even if exchangeability condition fails:(conservative) error control for E[V ] holds
Motif regression (n=287, p=195)
I two stable motifs with stability selection (w.r.t. E[V ] ≤ 1)I Multi sample splitting finds only the one motif which is not
biologically validated
Ribolflavin data (n=71, p=4088): control for E[V ] ≤ 1
●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●
●
●●●●●●●●
●
●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●
●
●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●
●
●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●
●
●
●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●
●
●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●
●
●
●●
●
●
●
●
●●●●●●●●
●
●●
●
●●●●●
●
●●
●
●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●●●●●
●
●●●●●●●●●●
●
●●
●
●●●●●●●●●●●
●
●●●●●●●
●
●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●
●
●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●
●
●
●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●
●
●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●
●
●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●
●
●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●
●
●
●
●●●●●●●●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●●●●●●●●●●●●●
0 1000 2000 3000 4000
0.0
0.1
0.2
0.3
0.4
0.5
0.6
stability selection E[V] <= 1
Index
rela
tive
sel.
freq
.
I stability selection: 4 stable variables (w.r.t. E[V ] ≤ 1)I Multi sample splitting: 2 significant variables (w.r.t. FWER)
Graphical modeling using GLasso(Rothman, Bickel, Levina & Zhu, 2008; Friedman, Hastie & Tibshirani, 2008)
infer conditional independence graph using `1-penalizationi.e. infer zeroes of Σ−1 from X1, . . . ,Xn i.i.d. ∼ Np(0,Σ)
Σ−1jk 6= 0 ⇔ X (j) 6⊥ X (k)|X ({1,...,p}\{j,k}) ⇔ edge j − k
gene expr. data
?⇒
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.46
Gra
phic
al Lasso
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
Sta
bili
ty S
ele
ction
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.448
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.436
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.424
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.412
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
!! == 0.4
!!
!
!!
!!
!
!!!!!!
!
!!!!
!
!!!!!!!!
!
!
!
!!
!!!
!
!
!
!!!
!!!!
!
!
!!!!
!!
!!!
!
!!
!
!!!!
!
!!!
!
!
!
!!
!
!!!!
!
!!!
!
!!
!
!!!!
!
!
!!!!!
!
!!!!!!
!
!!!
!
!!
!
!!!
!
!!!!
!
!
!
!!!!!!!!
!
!
!
!
!
!!!
!
!
!!!!
!
!!!!
!
!!
!!!
!!!
zero-pattern of Σ−1
sub-problem of Riboflavin datap = 160, n = 115stability selection with E[V ] ≤ 5
varying the regularization parameter λ in `1-penalization
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.46
Gra
phic
al L
asso
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
Sta
bilit
y S
elec
tion
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.448
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.436
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.424
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.412
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
λλ == 0.4
●●
●
●●
●●
●
●●●●●●
●
●●●●
●
●●●●●●●●
●
●
●
●●
●●●
●
●
●
●●
●●●●●
●
●
●●●●
●●
●●●
●
●●
●
●●●●
●
●●●
●
●
●
●●
●
●●●●
●
●●●
●
●●
●
●●●●
●
●
●●●●
●
●
●●●●●●
●
●●●
●
●●
●
●●●
●
●●●●
●
●
●
●●●●●●●●
●
●
●
●
●
●●●
●
●
●●●●
●
●●●●
●
●●
●●●
●●●
with stability selection: choice of initial λ-tuning parameter doesnot matter much (as proved by our theory)just need to fix the finite-sample control
permutation of variablesvarying the regularization parameter for the null-case
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.065G
raph
ical
Las
so
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
Sta
bilit
y S
elec
tion
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.063
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.061
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.059
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.057
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.055
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
with stability selection: the number of false positives is indeedcontrolled (as proved by our theory)and here: exchangeability condition holds
permutation of variablesvarying the regularization parameter for the null-case
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.065G
raph
ical
Las
so
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
Sta
bilit
y S
elec
tion
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.063
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.061
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.059
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.057
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●
●●●●
●●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
λλ == 0.055
●●
●●●●●
●
●●●
●●●●
●●
●●
●
●●●●●●
●
●●●●●●●●●●●●●
●●●●
●●●
●
●
●●●●
●●●●
●
●●●●●
●
●●
●
●●●
●
●●●
●
●
●
●●●
●
●●●●
●●
●
●
●
●●●●●●
●●●●
●●●
●
●●●●●
●
●●●●●●
●
●
●●
●
●●
●
●● ●●
●
●
●
●●●
●
●
●●●●●●●●
●●
●●
●
●●●●●●●●●●
●
with stability selection: the number of false positives is indeedcontrolled (as proved by our theory)and here: exchangeability condition holds
Predicting causal effects from observational data
↘ error rate
Causal gene ranking
summary median errorGene rank effect expression (PCER) name
1 AT2G45660 1 0.60 5.07 0.0017 AGL20 (SOC1)2 AT4G24010 2 0.61 5.69 0.0021 ATCSLG13 AT1G15520 2 0.58 5.42 0.0017 PDR124 AT3G02920 5 0.58 7.44 0.0024 replication protein-related5 AT5G43610 5 0.41 4.98 0.0101 ATSUC66 AT4G00650 7 0.48 5.56 0.0020 FRI7 AT1G24070 8 0.57 6.13 0.0026 ATCSLA108 AT1G19940 9 0.53 5.13 0.0019 AtGH9B59 AT3G61170 9 0.51 5.12 0.0034 protein coding
10 AT1G32375 10 0.54 5.21 0.0031 protein coding11 AT2G15320 10 0.50 5.57 0.0027 protein coding12 AT2G28120 10 0.49 6.45 0.0026 protein coding13 AT2G16510 13 0.50 10.7 0.0023 AVAP514 AT3G14630 13 0.48 4.87 0.0039 CYP72A915 AT1G11800 15 0.51 6.97 0.0028 protein coding16 AT5G44800 16 0.32 6.55 0.0704 CHR417 AT3G50660 17 0.40 7.60 0.0059 DWF418 AT5G10140 19 0.30 10.3 0.0064 FLC19 AT1G24110 20 0.49 4.66 0.0059 peroxidase, putative20 AT1G27030 20 0.45 10.1 0.0059 unknown protein
• biological validation by gene knockout experiments in progress.
; see Friday...
The Lasso and its stability path and why Stability Selection works so well
riboflavin example: n = 71, p = 4099sparsity s0 “=” 6 (6 “relevant” genes;
all other variables permuted)Lasso Stability selection
with stability selection: the 4-6 “true” variables are sticking outmuch more clearly from noise covariates
stability selection cannot be reproduced by simply selecting theright penalty with Lasso
stability selection provides a fundamentally new solution
Comparative conclusions
without conditions on the “design matrix” or “covariancestructure” (etc.):cannot assign strengths or uncertainty of a variable or astructural component (e.g. edge in a graph) in ahigh-dimensional setting
and these conditions are typically uncheckable...
three “easy to use” methods in comparison:
method assumptions applicabilitymulti sample splitting compatibility condition GLMs
beta-min conditionstability selection exchangeability condition “all”
the less assumptions, the more trustable the statisticalinference!“yet”, given the necessity of often uncheckable “designconditions”:confirmatory high-dimensional inference remains challenging