Comprehensive miRNA expressionanalysis in peripheral blood can diagnose               liver disease          Yoshiki Murak...
liver disease:    Discrimination vs normal※Hepatocellular carcinoma                           (HCC, liver cancer)※Chronic ...
Task under the following Conditions:※Small number of samples (patients).Less than 100, sometimes <10 in each class※High re...
What is miRNA?・miRNA = microRNA・non-coding RNA・regulate (suppress) target gene expression via eitherdegradation or transla...
What is exosome?
※Why blood?→non-invasive, less stress to patients※Why miRNA?→possibility to diagnose multiplediseases with miRNAs measurem...
Number of samples (patients)             Training        Test    CHB             4             16    CHC            64    ...
Method:PCA-based Linear discriminant analysis withPCA-based feature extraction                  without labeling informati...
Results for training data(Leave-one-out cross validation)                              Result                       CHB   ...
Embedding with PCA (2D)                          +: CHB,                          ◦: CHC,                          ×:NASH ...
NEED for increasing samples in silicoMCMCmiRNA 〜 age + gender + BMI +inflammation stage + fibrosis stageDistribution of co...
The Advantage of MCMC based resampling                      y=ax+by          X   X                                  x     ...
Embedding with PCA (2D)                          +: CHB,                          ◦: CHC,                          ×:NASH ...
Results for training data (LOOCV)Resampled                              Result                       CHB   CHC NASH   NL  ...
Test sample comes ….Requirement :※Use same miRNAs selected for Training dataAns.: Semi-supervised PCA-based LDA Training +...
result                     Training      resample       Test                    CHB NASH      CHB NASH    CHB NASH        ...
Conclusions※From clinical point of views, miRNAs inexsome in blood can diagnose liver diseases※PCA-based FE is useful one ...
BTW, It was press released....                  “We will try to make it                  recognized as a highly           ...
Upcoming SlideShare
Loading in …5
×

Comprehensive miRNA expression analysis in peripheral blood can diagnose liver disease

960 views

Published on

SIGBIO32

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
960
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
17
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Comprehensive miRNA expression analysis in peripheral blood can diagnose liver disease

  1. 1. Comprehensive miRNA expressionanalysis in peripheral blood can diagnose liver disease Yoshiki Murakami (M.D.), Hidenori Toyoda, Toshihito Tanahashi, Junko Tanaka,Takashi Kumada, Yusuke Yoshioka, Nobuyoshi Kosaka, Takahiro Ochiya, Y-h Taguchi PLoS ONE (2012) 7(10): e48366
  2. 2. liver disease: Discrimination vs normal※Hepatocellular carcinoma (HCC, liver cancer)※Chronic hepatitis C (CHC, liver inflammation type C )※Chronic hepatitis B (CHB, liver inflammation type B)※Non-alcoholic fatty liver disease (NAFLD), included Nonalcoholic steatohepatitis (NASH →Cirrhosis → HCC )
  3. 3. Task under the following Conditions:※Small number of samples (patients).Less than 100, sometimes <10 in each class※High reproducibility (for clinical application)※Feature extraction (FE) with stability(hopefully < 10 miRNAs among more than a fewhundreds miRNAs [in exosome in blood] )Cf. Compressive Sensing
  4. 4. What is miRNA?・miRNA = microRNA・non-coding RNA・regulate (suppress) target gene expression via eitherdegradation or translation interruptionof target genes
  5. 5. What is exosome?
  6. 6. ※Why blood?→non-invasive, less stress to patients※Why miRNA?→possibility to diagnose multiplediseases with miRNAs measurementsonly.(Cf. YT and YM, SIGBIO28 2012.3)
  7. 7. Number of samples (patients) Training Test CHB 4 16 CHC 64 31 NASH 12 8Normal(NL) 24 0887 miRNAs in exosome in blood(Agilent miRNA array)+ clinical information(age, sex, BMI, inflammation, fibrosis etc...)
  8. 8. Method:PCA-based Linear discriminant analysis withPCA-based feature extraction without labeling information(YT and YM, SIGBIO28, 2012. 3 at Tohoku U) because of stability requirementVery few researches of FE focusing stability(eg. Varshavsky R et al (2007) Bioinformatics)Unsupervised Feature Filtering (UFF)more computationally massive than ours
  9. 9. Results for training data(Leave-one-out cross validation) Result CHB CHC NASH NL CHB 2 0 1 2 Prediction CHC 0 64 1 3 NASH 1 0 9 3 NL 1 0 1 16Accuracy : 87.5% with 12 miRNAsup to 13 PCs (up to 16 probes for each miRNA)
  10. 10. Embedding with PCA (2D) +: CHB, ◦: CHC, ×:NASH △: NL.
  11. 11. NEED for increasing samples in silicoMCMCmiRNA 〜 age + gender + BMI +inflammation stage + fibrosis stageDistribution of coefficient ×Resampled clinical information =    miRNA
  12. 12. The Advantage of MCMC based resampling y=ax+by X X x a ↔ b : correlated
  13. 13. Embedding with PCA (2D) +: CHB, ◦: CHC, ×:NASH △: NL.
  14. 14. Results for training data (LOOCV)Resampled Result CHB CHC NASH NL CHB 94 0 0 6 Prediction CHC 0 99 1 0 NASH 6 1 97 3 NL 0 0 2 91Accuracy : 95.25% with 12 miRNAsup to 5 PCs (up to 16 probes for each miRNA)
  15. 15. Test sample comes ….Requirement :※Use same miRNAs selected for Training dataAns.: Semi-supervised PCA-based LDA Training + Test → PCA PCA-based LDA with Training sample label → discriminate Test Set※Validate in silico resmpling by independent dataAns.: validated by HCC data (not shown here)
  16. 16. result Training resample Test CHB NASH CHB NASH CHB NASH CHB 3 1 91 0 11 0 NASH 1 11 9 100 5 8Prediction Training resample Test CHB CHC CHB CHC CHB CHC CHB 3 0 100 0 15 11 CHC 1 64 0 100 1 20 Training resample Test CHC NASH CHC NASH CHB NASH CHC 64 2 100 0 29 3 NASH 1 10 0 100 2 5
  17. 17. Conclusions※From clinical point of views, miRNAs inexsome in blood can diagnose liver diseases※PCA-based FE is useful one as a unsupervised FE※MCMC based in silico resampling is useful (only for miRNA?)※semi-supervised PCA-based LDA can work (only for miRNA?)
  18. 18. BTW, It was press released.... “We will try to make it recognized as a highly advanced medical treatment in three years” (!)

×