Assump ion-awa e in insic molecula sub yping
o b eas cance esea ch
Qiao Yang1, Johan Ha man1,2, Emmanouil G. Si akis1
1Depa men o Oncology-Pa hology, Ka olinska Ins i u e , S ockholm, Sweden
2Depa men o Clinical Pa hology and Cance Diagnos ics, Ka olinska Uni e si y Hospi al
In oduc ion
•In insic molecula sub ypes a e associa ed wi h p ognosis
and suppo sys emic he apy decisions in b eas cance .
•In clinical p ac ice, sub yping is s anda dised; in esea ch,
a iable p ep ocessing pipelines and unme coho
assump ions can shi calls.
•A uni ied, assump ion-awa e amewo k is needed o
imp o e eliabili y and ep oducibili y in esea ch.
Con ac
Emmanouil G. Si akis, PhD
Ka olinska Ins i u e
emmanouil.si [email protected]
Objec i e
•P o ide a uni ied, assump ion-awa e Bioconduc o
amewo k o mo e accu a e, ep oducible in insic
molecula sub yping in b eas cance esea ch.
Key esul s
•Skewed coho s: AUTO imp o es accu acy by 18–19
pe cen age poin s and Cohen’s kappa (κ) by +0.20–0.26 s
Excluded (by AUTO) (Fig. 2).
•Empi ical h esholds: O iginal NC me hods become un eliable
ou side 39%–69% ER+ o n is small; AUTO excludes
acco dingly (Fig. 3).
•QC: En opy lags low-con idence calls, suppo ing cau ious
in e p e a ion and ep oducibili y (Fig. 4).
Fig. 1 AUTO logic: only un me hods when coho assump ions
(ER/HER2 balance, sub ype pu i y, subg oup size) a e me .
Fig. 2 Skewed coho s (S10/S90): (A) S10 = 10% ER+; (B) S90 = 90%
ER+. AUTO ou pe o ms Excluded (by AUTO) in κ. Blue = AUTO-
enabled; O ange = Excluded (by AUTO).
Fig. 4 En opy QC: Highe en opy = mo e
unce ain y. QC lags low-con idence in insic
sub ype calls o guide cau ious in e p e a ion.
Fig. 3 Th esholds (NC): (A) ER+ p opo ion, un eliable ou side 39–
69%. (B) ER+ subg oup size: n < 15. (C) ER–subg oup size: n < 18.
AUTO excludes o iginal NC me hods unde hese condi ions.
A B C
A B
Resou ces
B eas Sub ypeR
JohanHa manG oupBio eam/B eas Sub ypeR Scan o
package + docs
(Bioconduc o )
Conclusions
•Uni ied oolki o ep oducible in insic molecula sub yping in
esea ch.
•AUTO educes misuse in skewed/small coho s and
imp o es accu acy.
•A ailable ia Bioconduc o and Shiny app.
B eas Sub ypeR p o ides a uni ied, assump ion-awa e
amewo k ha makes in insic molecula sub yping in
b eas cance esea ch mo e accu a e and ep oducible.
Me hods
•Classi ie s: Nea es -Cen oid (NC) and Single-Sample
P edic o (SSP) me hods ia one in e ace.
•AUTO mode: sc eens ER/HER2 balance, sub ype pu i y, and
subg oup size; enables only me hods whose assump ions a e
me (Fig. 1).
•No malisa ion and gene mapping: me hod-speci ic de aul s
wi h En ez-ID mapping.
•Ou pu s: sub ype calls, en opy-based QC, c oss-me hod
conco dance, and buil -in plo s.
