Transcript 投影片下載 - 資訊科學與工程學系

From Laboratory to Hospital – The
New Challenge of Bioinformatics
Researchers
唐傳義
清大資訊系
[email protected]
合作醫院及相關項目
•癌症
林口長庚口腔癌團隊
林口長庚婦癌團隊
工研院生醫中心(成大醫院食道癌團隊、
嘉義基督教醫院團隊)
•感染性疾病
長庚病毒中心(流感病毒、腸病毒)
署立新竹醫院與竹東榮民醫院(孢氏不動桿菌)
清大生命科學系(幽門桿菌、白色念珠球菌)
由演講、合開課程及共同執行計畫建立合作關係
從講述「淘汰低產蛋能力雞的經驗」開始
•收集台灣土雞不同發育時期之四種蛋白質
樣品，根據實驗室分析結果，判讀蛋白質
樣品及其濃度
•記錄每隻台灣土雞之產蛋數量
•基於蛋白質樣品濃度及蛋產率，設計篩選
法提昇台灣土雞產蛋率
•目前已找出方法可在14週就可預測雞的未
來是否為低產雞，專利申請中
•合作:動物科技研究所及雞場
Egg production rate of TRFCC (n=157).
(A) Total egg number of all hens, (B) hens in four
groups
(A)
(B)
100
120
Group I
Group II
Group III
Group IV
90
Egg production rate (%)
Egg production rate (%)
100
80
60
40
80
70
60
50
40
30
20
20
10
0
0
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
week of age
25
30
35
Week of age
40
45
Lack of association of relative protein levels
with total egg numbers
(A) Vitellogenin
(B) Apo A-I
6
24w (r=0.14)
35w (r=-0.52. p<0.01)
24w (r=0.23, p<0.01)
35w (r=0.53, p<0.01)
5
5
Relative levels of apo A-I
Relative levels of vitellogenin
6
4
3
2
1
4
3
2
1
0
0
30
50
70
90
Total egg number
110
130
30
50
70
90
Total egg number
110
130
組合序號篩選法 (幾何上鄰近點問題)
•將兩批雞的4個蛋白質濃度轉成序號(Rank),從一
批已知低產雞的蛋白質濃度可以找出其序號組合
碼，利用序號轉換方式搜尋出另一批雞有類似序
號組合碼的雞,預測其為低產雞
•檢查14週血清蛋白質序號組合碼 (尚未產蛋),即
已可以發現其與未來低產雞的強烈規則性
•利用組合序號篩選法可於14wk淘汰19.5%雞隻，其
中包含78.8% 之50%低產雞
臨床醫療資訊探勘與轉譯醫學
授課教師：唐傳義教授、統計學研究所謝文萍助理教授、核
子醫學科閻紫宸主任、婦癌研究中心賴瓊慧主任、神經科
學研究中心陸清松主任、腦中風中心李宗海主任
上課時間：S2S3
課程說明
本課程將從臨床醫學、醫療統計與生物資訊的整合性角
度，探討在後基因體時代如何進行臨床醫療資訊探勘與轉
譯醫學的發展研究。此課程著重於結合臨床電子病歷資訊，
臨床醫學圖像資訊，癌症臨床醫學及遺傳性疾病之基因庫
分析。本課程集合來自臨床醫師、資訊、生物統計教授等
師資，以跨領域研究實做專題方式，引入生物統計與系統
生物學相關資訊技術，將珍貴的臨床醫學資訊作加值應用，
以輔助制訂更有效的醫療決策模式。
Cdc2 cutoff 1
NDRG1 cutoff 1
EF1A
cutoff 2
Biomarker Visualization: Functional
Interaction Linkage Map
11,612
11,402
353,043
6,434
SNP
34,435
169,093
59,852
594,111
……….
dbEST
9
KIT
FLT1
KDR
ERBB
2
EGFR
ESR1
PGR
1
0.107
0.142
0.137
0.084
0.022
0.01
FLT1
1
0.268
0.102
0.055
0.016
0
KDR
1
0.102
0.069
0.027
0.024
ERBB
2
1
0.231
0.043
0.026
EGFR
1
0.057
0.026
ESR1
1
0.088
PGR
1
KIT--- C-KIT,
FLT1--- VEGFR-1,
KIT
KDR --- VEGFR-2
ERBB2 --- HER-2/neu
EGFR --- EGFR
ESR1 --- ER
PGR --- PR
the ratio of common neigh
長清計畫
Oral Cancer
with 閻紫宸
Imaging and
Clinical Data
Psycho-social
Study and
Supportive
Care
Epidemiology
and
Translational
Study
互補、信賴
Informatics
Systems Biology Approach: 4 M’s
Paradigm
清
華
團
隊
長
庚
團
隊
GRP78 knockdown inhibits cell invasion
(A) FADU
(B) Detroit
Scramble
Scramble
siRNA-1
800
200
150
100
50
*
P=0.016
0
Vector
Vector
siRNA-1
Number of invaded cells
Number of invaded cells
Vector
600
400
200
P=0.0013
**
0
Scramble
siRNA
Vector
Scramble
siRNA
Systematic analyses flow:
Survival analyses
& data mining
Biomarkers
Map to FILM
Functional module
finding
Expression profiles
Significant changed genes
Map to FILM
Clustering
Up- or downregulation
Map to KEGG
Pathways finding
Disease network
Map to FILM
Pathway cross linking
Drug-target network
14
Pathway prediction
Missing link
Missing gene
15
hsa7153
hsa2099
has6256
16
長庚頭頸癌研究團隊
病
歷
資
料
長庚婦癌研究團隊
長庚癌症臨床
及研究中心
剩餘檢體
長庚大學生物技術暨檢驗學系
血清免疫
生物標記測量
長庚臨床病理科
資料探勘,
存活分析,
系統生物分析
其他生物標記測量
長庚大學生化與生醫工程系
基因晶片測量
清華大學資訊工程系
Sample Classification
pN+
There are
112 patients
in total.
Relapse
Tumor depth
ECS+
SUVnodal >= 5.7
20
ECS+
SUVnodal < 5.7
14
ECS-
SUVnodal >= 4.1
14
ECS-
SUVnodal < 4.1
14
<= 10 months
SUVnodal >= 5.5
11
<= 10 months
SUVnodal < 5.5
15
> 10 months
SUVnodal >= 3.6
10
> 10 months
SUVnodal < 3.6
13
>= 12mm
SUVtumor >= 19.3
13
>= 12mm
SUVtumor < 19.3
40
<= 12mm
SUVtumor >= 19.3
12
Recurrence
11
Second Primary
10
pT4N0
17
pT1N2b
12
NOABC
11
18
Data provided by CGMH:
Analysis Plan
Phenotype
Differentially
expressed genes
Association
Genotype
(CNVs, SNPs)
Expression
Genetic
components
of expression
20
Biomarker Finding Plan
DNA loci
Causal genes
L1
g1
L2
L3
g2
g3
L4
……….
……….
Rn
W2
W2
W3
g4
W5
g5
Wn
Traits I
W3
W4
g2
g3
g4
W5
Wn
W1
W2
W3
W4
W5
g5
……….
R2
W1
W1
……….
R1
g1
W4
Ln
Reactive genes
gn
gn
Wn
Traits
II
SNP 6.0
Association
study
LOH
analysis
Exon array
Copy
number
analysis
Biological
network
analysis
QTL
analysis
Biomarkers at DNA/ mRNA level
Tissue
array
Biomarkers at DNA level
Diagnosis chip design
Biomarkers at protein level
臨床病歷
及生醫資料庫
疾病關連與
藥物-靶點作用
網路資料庫
FILM資料庫
(基因功能
網路)
臨床病歷及
生醫資料分析
平台
Gene Expression
分析模組
Survival
分析模組
SNP Array
分析模組
分析平台
生醫資料庫
Text Mining
模組
Gene Module
Analysis模組
Pathway
Analysis模組
Biomarker and
Drug Target
Prediction 模組
臨床知識探勘平台
New Module:
Next-generation Sequencing
Data Analysis
• Roche 454 GS-FLX System
• ABI SOLID sequencing system
• Illumina Solexa 1G Genome Analyzer
Read: short (35~100bp), small error rate, high
coverage and low cost.
Genome
Reads
Re-sequencing Problem Definition
• We are given a text T=t1t2…tn, a set of
patterns P1, P2, … PN , and a constant k.
We are asked to find all the occurrences of
Pj in T with k errors (Hamming distance).
Algorithms
• Indexing Genome with Hash Tables:
SOAP …
• Indexing Reads with Hash Tables: MAQ,
ZOOM, SeqMap and RMAP, …
• Indexing Genome with Suffix
Array/BWT: Bowtie, …
Indexing Genome with Suffix Array/BWT
• Bowtie algorithm is the faster one.
This is probably the fastest short read aligner to date.
Length
36bp
Program
CPU time
Bowtie
6 m 15 s
Maq
3 h 52 m 54 s
SOAP
16 h 44 m 3 s
As Quick as Bowtie and with Ability of
Alignment Distance
• If there are many indels (deletions or insertion)
when align sequencing data onto reference
sequence, the results of alignment with
Hamming Distance are not acceptable. (<40% of
read mapped)
Delete
…ACGGATAGCTAGCTAGCATCAGGGCAGATCA…
TAGCTAGCTGCATCAGGG
GCTAGCTGCATCAGGGCA
AGCTGCATCAGGGCAGAT
+G
Insert
…ACGGATAGCTAGCTCATCAGGGCAGATCA…
TAGCTAGCTGCATCAGGG
GCTAGCTGCATCAGGGCA
AGCTGCATCAGGGCAGAT
WorkFlow
NGS data
Filtering
Mapping Algorithm
(Bowtie) (Trim 17 bps
and set error bound=2)
Mark Mapped Regions
(depth=5)
Progressive Mapping
Trim 3 bps
Unmapped
Reads
Yes
No
SNP found?
Map reads to marked regions
with Hamming distance =1
Score and modify the
ref-sequence
hsa-miR-99a
59
2
IGF1R(3480)
hsa-miR-99a
target
SMARCD1
(6602)
13
78
WISP2
(8839)
HADHB
(3032)
Glucocorticoid receptor regulatory network(NCI)
Beta oxidation of palmitoyl-CoA to myristoyl-CoA(Reactome)
mitochondrial fatty acid beta-oxidation of unsaturated fatty acids(Reactome)
Fatty acid elongation in mitochondria(KEGG)
IGF1 pathway(NCI)
E-cadherin signaling events(NCI)
Plasma membrane estrogen receptor signaling(NCI)
Integrins in angiogenesis(NCI)
Green: miRNA
target
Red: mRNA
Purple: miRNA +mRNA
target
感染科醫生常問的問題
• 為新的本土抗藥菌株定序 (NGS Assembly)
• 了解抗藥菌株的抗藥機置
(NGS re-sequencing, SNP finding)
• 利用過去資料找宿主專一性，毒性
• 設計新藥
HP Experiments
(Edit distance)
Reference sequence size ≒1.6M
Read Number: 3074139
Read Size = 76bp
% of read mapped
Ref
1
Ref
2
Ref
3
Ref
4
Ref
5
Ref
6
Ref
7
error
59% 59% 55% 57% 50% 57% 67%
<6
Genome Sequence Modification
Ref Seq: …
A
C
C
G
A
T
C
A score
25
0
1
0
1
0
1
C score
1
30
35
0
40
1
32
G score
0
1
0
40
0
0
0
T score
1
0
0
0
0
1
0
Deletion
score
0
0
0
0
0
30
0
Insertion
score
+A
31
Modified sequence: …ACACGCTC…
…
WorkFlow
Short read data
Map reads to ref-sequence
with small edit distance
Map reads to Modified refsequence with small edit
distance
Score and output
modified ref-sequence
Yes
Indel
found?
No
End
error < % of read
6
mapped
# of
deletion
# of
insertion
Genome
Coverage
Run 1
61.6%
497
435
86.6%
Run 2
71.2%
235
234
88.9%
Run 3
72.7%
132
157
89.6%
Run 4
73.2%
78
82
90%
Run 5
73.5%
37
74
90.2%
73.9%
13
9
90.4%
…
Run 16
找功能所需要的特殊軟體：以 RNase proteins 為例
PI
RNase 1
(human pancreatic rnase)
(HPR)
9.1
RNase2
(eosinophil-derived neurotoxin)
(EDN)
9.2
RNase 3
(eosinophil cationic protein)
(ECP)
(potent anti-parasitic agent)
10.8
RNase 4
9.3
RNase 5 (angiogenin) (ANG)
9.73
RNase 6 (k6)
9.49
RNase 7
10.5
RNase 8
8.6
substrate
specificity
Antiviral activity
Antibacterial
activity for
prokaryote
neurotoxicity
No
No
No
Yes
Yes (E. coli)
Yes
(required active
ribonuclease
activity)
Yes
(against RSV)
Yes (E. coli)
Yes
(much lower
than EDN)
uridine-preferring
No
No
No
tRNA-specific
cytidine preferring
No
No
No
No
No
unknown
No
Yes
unknown
No
No
Viral genomic
DNA
(antiviral activity against
RSV and HIV studied in
vitro)
Multiple Sequence Alignment
• Given s set of sequences,the MSA problem
is to find an alignment of the sequences
such that some object function is
minimized
• ie.(Sum of Pair Score)
S1：ATTCG
S2：AGTCG
S3：ATCAG
S’1：A T – T C – G
MSA
S’2：A – G T C – G
S’
3：A
T–– CAG
Cost = 8
2
4
2
MSA of 7 RNases
(by Workbench3.2)
MSA of 7 RNases
1ONC
1RCN
1DE3
1BC4
PDB ID
1ONC
1RCN
1DE3
1BC4
RNase
Name
Source
Pancreatic Ribonuclease
Ribonuclease A
Ribonuclease -Sarcin
Ribonuclease
Rana Pipiens
Bovine (Bos Taurus)
Pancreas
Aspergillus Giganteus
Rana Catesbeiana
Recombinant/Native
pI
Native
Native
Recombinant
Native
8.96
8.64
9.17
9.20
disulfide
bond
4
4
2
4
Substrate/li
gand
SO4
Dna (5'-D(Aptpapap)-3')
X-ray/NMR
X-ray
X-ray
NMR
NMR
J Biol Chem 269 pp. 21526
(1994)
J.Mol.Biol. 299 pp. 1061
(2000)
J Mol Biol 283 pp. 231
(1998)
Reference
J Mol Biol 236 pp. 1141
(1994)
Multiple Sequence
Alignment with
Constraints
MuSiC (Bioinformatics, 2004)
MuSiC-ME (Memory Efficient, Bioinformatics,
2005)
RE-MuSiC (Regular Expression, NAR, 2006)
Multiple Sequence Alignment
with Constraint
• Input: (1) multiple Protein/DNA/RNA
sequences and (2) several constraints
(represented by regular expressions), with
each consisting of known functionally,
structurally or evolutionarily related
residues/nucleotides of the input sequences.
• Output: an optimal multiple sequence
alignment in the condition that the constrained
amino acids/ nucleotides should be aligned
together in the alignment.
CMSA: Constrained Multiple Sequence
Alignment Problem
• Input: a set of k sequences along with a order set
of r constraints (C1, …, Cr) and an error ratio 0  
<1
• Output: an optimal CMSA, say A, in which r
disjoint bands B1, …, Br are in A such that
d(Ci, Bi(Sj))  l(Ci) for all 1ir and 1jk.
–
–
–
–
band: a block of consecutive columns in A
d(x,y): the Hamming distance between x and y
Bi(Sj): the fragment of Sj whose bases are all in Bi
Ci: the length of Ci (also denoted by i)
Example of CMSA
• Input: 6 RNA sequences along with 11
constraints and error ratio =0
Example of CMSA
• Input: 6 RNA sequences along with 11
constraints and error ratio =0.5
Web Interface of MuSiC
http://genome.life.nctu.edu.tw/MUSIC/
Syntax of Regular Expression
• The IUPAC codes for the amino acids and
nucleotides are used in the regular expression.
• “-”: separate the elements of a regular
expression.
• “[]”: the amino acids (or nucleotides) that are
allowed to appear at a given position.
• “{}”: The amino acids (or nucleotides) that are
not accepted at a given position.
• Repetition of an element is indicated by
appending, immediately following that element,
an integer or a pair of integers in parentheses.
Example: G-[AG]-x(4)-{AG}-x(4,5)
RE-MuSiC: Multiple Sequence Alignment
with Regular Expression Constraints
限制型多重序列比對的軟體工具
RE-MuSiC發表在Nucleic Acids Research (Vol. 35, pp. W639-644,2007)
Too many false
positives !
MSA of RNase1~RNase6
- 3 active sites
(His42, Lys65, His155 )
- 4 disulfide bonds
(Cys50, Cys64, Cys82, Cys89,
Cys98, Cys110, Cys123, Cys138 )
- MSA showed that 11 residue
were conserved in RNase3,
RNase2 (functionally related
enzyme) and RNase1~RNase4
(sequence related proteins).
Clustering of 8 RNase
• Group1: ECP (RNase3)
• Group3: EDN (RNase2) that is functionally related to ECP
• Group2: RNase1, RNase4, RNase5 and RNase6 don’t
have toxicity.
• Group4: RNase7 and RNase8 have toxicity, but their
toxicity is still unknown.
Algorithm
Voting score (1)
Rat imidase
Aye from G3
Blackball from G2
Vscore
There are five possible coordinates: (1) Residues at rat imidase,
functionally identical or related proteins (group3 or group4, respectively) and sequence related
proteins (group2) are different, the score is set to zero. (2) The score is set to 1 if residues at all
sequences are the same. (3) Residues common at rat imidase and proteins of group3 or group4 but
differ from that of group2, the score is set to 3. (4) Residues common at imidase and group2
proteins but differ from that of group3 or group4, the score is set to –2. (5) Residues common
at sequence related proteins and functional related proteins but differ from that of imidase,
the score is set to zero.
RNase:
Comparison of MSA and
our method (2)
- The first row is the amino
acid sequence of ECP, the
second and the third row
represent the total scores and
their correspondent ranks
respectively.
- green residues: the top 5
high ones in our method
- red residues: 3 active sites
and 4 disulfide bonds of RNase
proteins
- Pro3 was verified to be
associated with ECP’s toxicity
by biological experiments.
FAVFAT
• Revealing the desired features of target enzyme or protein by
voting on three different property groups aligned by threeprofile alignment method. (accepted by BMC genomic 2010)
• Three properties
– Target (interested sequence)
– Property A (related function sequences)
– Property ~A (Non-function sequences)
• Goal :
– Identifying amino acid residues critical for Human
Enterovirus 71.
– Identifying function and species-associated sites for
Influenza A virus
Schematic diagram of the influenza virus replication cycle
Our approach
• 3D-QSAR (Pharmacophore) model design
• Chemical compound inference
• Drug synthesis and validation
59
2015/4/9
Structure of Neuraminidase protein
Influenza A virus Resolution：2.5Å
Sialic acid
Active Site
Pharmacophore Generation
A series of inhibitors
Comformations generation
Pharmacophore
( Hypothesis )
6
1
Drug Screening
Synthetic compounds
Natural compounds
~ 5,000,000 cpds
Database
~ 90,000 cpds
Build Feature Model
OH
F
HY
O
S
O
OH
OH
N
OH
O
O
H 2N
O
S
HN
O
HN
NH
O
NH 2
HBA
N
O
OH
OH
N
OH
O
OH
O
HO
O
OH
NH
OH
N
H
HN
HN
NH
O
NH
O
F
F
F
NH 2
Training inhibitors for
feature model of
protein X
NH 2
RA
HBA
X’s Spatial Feature
X’s Spatial Feature Search
X’s Spatial recognition
Inhibitor candidates of Protein X
Chemical Compound Inference Problem
 Fujiwara et al. proposed a sequential
branch-and-bound algorithm to solve this
problem.
H. Fujiwara, J. Wang, L. Zhao, H.
Nagamochi, and T. Akutsu,
“Enumerating Treelike Chemical
Graphs with Given Path Frequency”, J.
Chem. Inf. Model., 2008, 48(7), pp.
1345-1357.
 The algorithm proposed by Fujiwara
cannot deal with the ring structure of
chemical compounds. Moreover, the
computation time increase significantly
when the number of atoms grows.
 In this study, we proposed a Balanced
Multi-Process Parallel Algorithm for
Chemical Compound Inference Problem.
BMPBB-CCI
• Balanced Multi-Process Branch-andBound Algorithm for Chemical Compound
Inference Problem
• The goal of BMPBB-CCI include
–Reduce computation time via parallel
computing
–Take care of ring structure of CCI problem
64
2009.08.21
66
2015/4/9
未來的新方向
• Mata Genomics
NGS analysis
GPU solution
• Cancer Genomics
SNP, Indel, Translocation detection
• Experiment Design
Introduction (Penn State project)
Here, we illustrate a scenario of microbial community
profiling.
Fig. 1. The scenario of collecting samples from a car and
the sequencing process.
Windshield Genomics
Sources
Why GPU?
Massively Parallel Processor
A quiet revolution and potential build-up
– Calculation: 367 GFLOPS vs. 32 GFLOPS
– Memory Bandwidth: 86.4 GB/s vs. 8.4 GB/s
– Until last year, programmed through graphics API
GFLOPS
•
G80 = GeForce 8800 GTX
G71 = GeForce 7900 GTX
G70 = GeForce 7800 GTX
NV40 = GeForce 6800 Ultra
NV35 = GeForce FX 5950 Ultra
NV30 = GeForce FX 5800
–
71
GPU in every PC and workstation – massive volume and
potential impact
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007
ECE 498AL1, University of Illinois, Urbana-Champaign
2015/4/9
GPU
Genome Rearrangements
and Evolutionary Trees
ROBIN (Bioinformatics, 2005)
SPRING (Nucleic Acids Research, 2006)
Genome
Rearrangements
區段互換的基因體重組
Human X
Mouse X
4
6
1
7
2
3
5
8
4
6
7
1
2
3
5
8
4
1
2
3
5
6
7
8
1
2
3
4
5
6
7
8
三種常見人類致病性弧菌的演化關係
Chromosome 1
Chromosome 2
創傷 腸炎 霍亂
弧菌 弧菌 弧菌
創傷 腸炎 霍亂
弧菌 弧菌 弧菌
創傷
弧菌
-
39
69
創傷
弧菌
-
3
6
腸炎
弧菌
39
-
65
腸炎
弧菌
3
-
7
霍亂
弧菌
69
65
-
霍亂
弧菌
6
7
-
研究成果發表在 J. Computational Biology, (Vol. 12, pp. 102-112. , 2005)
用反向工程技術
做蘭花花型基因探勘
實驗工具
RNAi
RNAi (RNA interference)
dsRNA被細胞雙鏈RNA特異的核酸酶切
成21-23個鹼基對的短雙鏈RNA
稱為 siRNA（small interfering RNA）
siRNA與細胞某些酶和
蛋白質形成複合體，稱
為RNA誘導沉默複合體
（RNA-induced silencing
complex,RISC）
RISC 可識別與siRNA有同源序列的mRNA
且在特異的位點將該mRNA切斷
開花功能探勘：蘭花基因工程
• 藉由載入與目標基因有同源序列的小片段雙股 RNA 誘發
RNAi 機制來達到抑目標基因表現的效果，做為探究基因
功能之新工具。
• 若載入的小片段雙股 RNA 與多條基因的片段有同源性，
則可以一次抑制多個基因的表現。
• 藉由分析蘭花基因序列，找出可以一次抑制多個基因表現
的可能雙股 RNA 序列。
• 使用挑選出的雙股 RNA 序列，在蘭花上進行 RNAi 實驗，
觀察產生變化之性徵，快速縮小與該性徵有關之可能基因
的範圍。
• 對已經篩選過的可能基因做第二次續列分析，重複 RANi
實驗，直到目標基因的個數減少至可以一一檢測的範圍。
According to similarity, find the center sequences and
determine its own group
• S1 is the center of a group G if S1 has no second neighbor
(Sec_nei_num(S1) =0).
S2
S3
G(S1) = {S1, S2, S3, S4 , S5}
S1
S5
S4
• If exist subsequence F, and HD(F,S1)=5,then F is a
Far_neighor (5) of S1.
F
S2
4
4
S5
S3
5
4
S1
4
S4
No.13
No.15
No.21
No.194
No.171
PR1 relative
7700 genes
TF No.1 …. No.272
siRNA from TF No.21
PR1
No.21
No.13
No.130
No.152
….… 146 ….…
No.74
No.112
No.168
No.176
Level 1
Level 2
No.21
No.13
PR1
PR1
No.130
No.152
7700 genes
TF No.1 …. No.272
siRNA from TF No.13
No.13
PR1
siRNA from TF No.21
siRNA from TF No.176
No.21
No.13
No.130
No.152
No.74
No.112
No.168
No.176
PR1
PR1 ---
120
100
120
120
80
100
100
60
80
80
40
60
60
20
40
40
0
20
20
0
0
感謝
•
•
•
•
•
•
•
•
•
•
•
•
實驗室全體成員
林口長庚口腔癌閻紫宸、廖俊達醫師團隊
長庚醫技鄭恩加教授實驗室團隊
林口長庚婦癌賴瓊慧醫師團隊
交大生資盧錦隆教授實驗室團隊
長庚資工林俊淵教授實驗室團隊
清大統計所謝文萍教授
清大生科王雯靜教授實驗室團隊
清大生科張大慈教授實驗室團隊
元培醫技劉明麗博士
台大植微葉信宏教授實驗室團隊
動物研究所李仁權博士