Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
コンペティションから見るAI創薬/AI drug discovery in the view ...
Search
m_mochizuki
March 18, 2019
Research
2
1.5k
コンペティションから見るAI創薬/AI drug discovery in the view of competitions
日本オミックス医学会シンポジウム 発表資料
場所: 東京医科歯科大学
日付: 2019/3/18
2018/3/20 誤記修正
2018/3/21 誤記修正
m_mochizuki
March 18, 2019
Tweet
Share
More Decks by m_mochizuki
See All by m_mochizuki
SIGNATE: 日本取引所グループ ファンダメンタルズ分析チャレンジ 1位解法 / the 1st place solution of JPX Fundamentals Analysis Challenge on SIGNATE
m_mochizuki
4
11k
SIGNATE: 日本取引所グループ ファンダメンタルズ分析チャレンジ 暫定1位解法 / the provisional 1st place solution of JPX Fundamentals Analysis Challenge on SIGNATE
m_mochizuki
3
9.4k
MD-DSC研究会講演資料:『機械学習コンペティションの実際とその意義』/ Practice on ML competition and its significance
m_mochizuki
1
1.2k
Other Decks in Research
See All in Research
能動適応的実験計画
masakat0
2
720
Computational OT #1 - Monge and Kantorovitch
gpeyre
0
220
cvpaper.challenge 10年の軌跡 / cvpaper.challenge a decade-long journey
gatheluck
1
270
Ad-DS Paper Circle #1
ykaneko1992
0
5.8k
大規模な2値整数計画問題に対する 効率的な重み付き局所探索法
mickey_kubo
1
310
20250725-bet-ai-day
cipepser
2
340
数理最適化に基づく制御
mickey_kubo
6
700
数理最適化と機械学習の融合
mickey_kubo
15
9.1k
SSII2025 [SS2] 横浜DeNAベイスターズの躍進を支えたAIプロダクト
ssii
PRO
7
3.8k
なめらかなシステムと運用維持の終わらぬ未来 / dicomo2025_coherently_fittable_system
monochromegane
0
1.6k
Agentic AIとMCPを利用したサービス作成入門
mickey_kubo
0
370
[CV勉強会@関東 CVPR2025] VLM自動運転model S4-Driver
shinkyoto
2
410
Featured
See All Featured
Principles of Awesome APIs and How to Build Them.
keavy
126
17k
Large-scale JavaScript Application Architecture
addyosmani
512
110k
jQuery: Nuts, Bolts and Bling
dougneiner
63
7.8k
Building an army of robots
kneath
306
45k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
29
9.6k
Docker and Python
trallard
45
3.5k
The Invisible Side of Design
smashingmag
301
51k
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
656
60k
Making the Leap to Tech Lead
cromwellryan
134
9.5k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
Java REST API Framework Comparison - PWX 2021
mraible
32
8.8k
Unsuck your backbone
ammeep
671
58k
Transcript
a I ( M A M89 :) 3 /21 1
/ 0/ ) ) A c
s ( 21 1- 40 z ( u 76 6
7 ) ) h o 7M M AI( : 7 r a ( k c 7 i
3 IS o J P 123 32 0 J P
T 0 J P T 0 J P T 6 n 0 J P T 4 DJ PG C A 7E5 D4J A C C eB E5
P GI gGI M ci 06?5:
0 65A 4? 5 B C 84 6 76 .?4 8 11ae h M ci 2 3 gM N M ci g T M
6 AI= +.9<*!%;7BG =?4D +.1:/"(&#*
5C0 A, "(&$ = %&' )6F5C=? >@8E3 -23
<; .%7 FC*
'4 <;p53L. + '&(AE!1IBM) DK5=(FC*'40 BH1G .%7D/:36I (-"8$) .%7>@1G JD ↑ # 2?, )9
1
1 0 0 n RAK ) S2 AK K n
g 1 0 Ra Ra 0 5 Q e ( 425 %10/7& (! * , 10/(! $%)+.# # " 425 'Kaggle(063(-
1 n eh a kGn f R / :/ V
Fod eh aFH lm g Sb RHMS A n V . :/ i eh a cS A p 2 / /: / ./: :.
M ,1 42 , 2 , 0 0 9 22
3. n ? n : AC
None
4 1 #)$ ' !%( * $
(Convolutional Neural Network) $ %" & &
1 $)& (" Fingerprint/Descriptor % & (Graph
Convolutional Neural Network) & ! # ' '
n 1 G 6 C n N : Altae-Tran et
al, ACS Cent. Sci.,2017,3(4), pp 283–293
n ( ) P O N B A n I
24 0 2 1 1 Virtual screening… 1 2 3
n N G n K C 967 ( #
&).0 *! "%*@ =1;=>* *(400) 4?:<A5).0 “”* */$, 862?3' +(-OK
7 8 T d fng ] N mi a M
hmi ] C [ Ct ep r a 21 . ? 0 ?9 5 ( ) ) ( )( 02 s y , 9 T u CG ] NaY I
( 7 2 Extended Connectivity Fingerprint Functional Connectivity Fingerprint Topological
Torsions Atom Pairs Fingerprint RDKit Fingerprint Avalon fingerprint !fingerprint (6) 70 Random Forest Extremely Randomized Trees Gradient Boosted Trees Multilayer perceptron Support Vector Regression ! $)%(&' (5) 65 = 30# Elastic Net Pfinal Level 0 Level 1 " # 1 2 ) ):
Fingerprint ECFP FCFP TT AP RDK AVLN F-Stacking RF
0.848 0.855 0.816 0.686 0.652 0.722 0.892 ERT 0.869 0.889 0.844 0.798 0.671 0.768 0.907 GBT 0.852 0.864 0.835 0.808 0.733 0.758 0.891 MLP 0.802 0.777 0.623 0.814 0.651 0.712 0.895 SVR 0.856 0.852 0.688 0.763 0.662 0.693 0.877 L-Stacking 0.890 0.911 0.870 0.881 0.799 0.846 0.930 FL-Stacking Level0 ROC-AUC ) 1 ) (0( 72 n 0 3 0 0
2 ( 7 )1)0 3 5
IMSBIO () ( ) 1 8 1 Univ-shizuoka 1 PFDrug ()Preferred Networks 1 kiharalab 1 1 Graph CNN
1 0
38 5 120 n 0 u ”Taklbe : () :.
00 1Tcn n s T n p w Th cn dg I I “ L ing n r P v T ng D n T t p o y 2 8: : .2 1 2 - // /: 0:
)0 3 6 ( 2 1 n Lel an f
LN b i - b i -/) - n U N gc d - s n D - b i ( 1) -1- P t ( 1) ) vo el an (
)0 37 ( 2 1 n 24 9: 9 0
4 n 1 9 9 56 8 2 3 W 2 9 4 O O O !!!
9 21 n e l ( g n K n
) a )
20 2 3 n 3 t 1 o r ru
2 a i 1 o ru ) 2 l1 ru 1es 2 1 r f K2 n (( h g g K2
1 3 n ( ) ) n : )
X 8 T 9 8. A T 9 8. A
T 9 8. A 0 3 T . 9? A 5 2
( • 9) : / 51) 5 • 1 55
5 2 1:023/ 5