15
A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY SCREENING OF CHEMICAL COMPOUNDS PI-111 Kohtaro Yuta In Silico Data Ltd. E-Mail : [email protected] http://www.insilicodata.com

A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

A NEW CLASSIFICATION METHODSUITABLE FOR TOXICITY SCREENING

OF CHEMICAL COMPOUNDS

PI-111

Kohtaro YutaIn Silico Data Ltd.

E-Mail : [email protected]://www.insilicodata.com

Page 2: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Why PR(Pattern Recognition)for toxicity screening

Toxicity1. Unexplained & complex

mechanisms2. High structural diversity

Factorial analysisCADD

Multi-variatePattern recognition

Artificial intelligence

Pred

iction

Black box

Know-how

Page 3: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Toxicity sample space :large overlapped space○ ×○ ○○○ ○

○ ○○

○ ○○○

×○○○

○ ○ ○

○ ○ ○ ○ ○○ ○ ○○○

○○ ○○

○○○○ ○○

○○○○ ○

○○○ ○ ○○○○

○ ○○○

○○

○ ○○

○ ○○○ ○

○○○○

○ ○ ○○

○○○

○○○

○○

○○

○○

○○

○○○○

○○○ ×× ×××

××××

×

××××

×

×

× ×× ×××

×× ×××

×× ×××××××× ×× × ××

××

××

× ×

×××

××××

× × ×××

××

× ××××

×××

× ×××× ××××××××

× ×××× ××

×

× ×

××

× ××××××× ××××

× ××× ×× ×

××

×××

××

× ×

○ ×○ ○○○ ○○ ○○

○ ○○

×○○○

○ ○ ○

○ ○○ ○ ○○ ○ ○○

○○

○ ○○

○○○○ ○○

○○○○ ○

○○○ ○ ○○○○

○ ○○○

○ ○

○ ○○

○ ○○○ ○

○○○○

○ ○○○

○○

○○

○○

○○

○○

○○

○○

○○○○

○ ○○ ×× ×××

××××

×

××××

×

×

× ×××××

×××

×××× ××

×××××× ×× × ×××

×

××

× ×

×××

××××

× × ×××

××

× ××××

×××

× ×××× ××

××

××× ×× ×××× ×

× ×

× ×

××

× ××××××× ××××

× ××× ×× ×

××

×× ×

×

×

× ×

○ ×○ ○○○ ○○ ○○

○ ○○

×○○○

○ ○ ○

○ ○○ ○ ○○ ○ ○○

○○

○ ○○

○○ ○○ ○○

○○○○ ○

○○○ ○ ○

○○○ ○○

○○

○ ○

○ ○

○○ ○○○ ○

○○○○

○ ○○○

○○

○○

○○

○○

○○

○○

○○○○○

○ ○○ ×× ×××

××××

×

××××

×

×

× ×××××

×××

×××× ××

×××××× ×× × ×××

×

××

× ×

×××

××××

× × ×××

××

× ××××

×××

× ×××× ××

××

××× ×× ×××× ×

× ×

× ×

××

×××

×××××

××

××

× ××× ×× ×

×× ×× ×

×

×

× ×

50% of samples

80% of samples

90% of samples

○ ×○○○○

○○

○ ○○

×

○○

○○ ○

○○

○○ ○

○ ○○○

○○

○○

○○○○

○○○

○○

○○

○○

○ ○

○○

○○○

○ ○○○

○○○

○○

○○

○○

○○

○○

○○

×× ×

××

××

××

×

××

××

×

×

× ×× ×

××

×× ×

××

××

×

×

××

××××

×× × ××

×

×

×

×

× ×

×

××

×

×××

××

×

×

×

×

×

× ××

×

×

××

×

× ×××× ×××

×

××

××

×××

×× ××

×

× ×

×

×

× ××

×

×

××

× ××

××

× ××

×

×

××

××

×

××

×

×

× ×

Normal sample space :small overlapped space

Page 4: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Perfect(100%) classification of Ames test 6965 pos/neg sample set

The most powerful and advanced data analysis method

The most difficult classification problem

6,965 sample of Ames test were,

Classified perfectly

K-step Yard sampling method KY-method

Page 5: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Classification Result by AdaBoost

Mis-classified region

Mis-classified region

77.24% of Ames test 6,965 samples

Correctly classified region

Correctly classified region

Page 6: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

○ ×○

○○

○○

○ ○○

×

○○

○ ○○

○○

○○

○○○○

○○

○○

○○

○ ○

○○

○○

○ ○○○

○○○

○○

○○

○○

○○

×× ×

××

××

××

×

××

××

×

×

××

× ×

×

×

×× ×

×

×

×

×

×

×

×

×

××××

×× × ×

×

×

×

×

×

× ×

×

×

×

×

××

×

××

×

×

×

×

×

× ××

×

×

××

×

××××××××

×

×

××

×

××

××

× ××

×

× ×

×

×

× ××

×

×

×

×

× ×

××

×

× ××

×

×

××

××

×

××

×

×

× ×

First basic concept of KY methodSpatial region on sample space

Both side of sample spacePure and no-overlapping on both region

Highly overlapped

TwoDiscriminant

functions Patent pended

Page 7: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Second basic concept of KY method

Multi-steps for 100% classification○ ×○

○○○ ○

○○

○ ○○

×

○○

○○

○ ○

○○

○○ ○○ ○ ○

○○

○○

○○

○○ ○○

○○○

○ ○

○○

○○○○

○ ○

○○

○○○

○ ○○○ ○

○○○

○ ○ ○○

○○

○○

○○

○○

○○

○○

○○

○○○○

○○

×× ×××

××××

×

××××

×

×

× ×× ×××

×× ×××

××

××

××

×××× ×× × ××

××

×

×× ×

×

××

×

×××

××

××

××

×

× ×××

×

××

×

× ×××× ××××

××

××

× ×××× ××

×

× ×

×

×× ××××

××

× ×××

×

× ×××

×× ×

××

×××

×

×× ×

Grey Zone50%

○ ×○○ ○

○○

○ ○○○

×

○○

○○○○

○ ○

○ ○○ ○○○

○○

○○

○○

○○ ○○

○○

○○○

○○○○

○○○

○ ○

○○

○○○ ○

○○○○

○○

○○

○○

○○○○

○××

×

××

×

××××

×

×

× ××

× ××

××

×

×××× ×× ××

××

×

× ×

×××

× ×× ×

××

××

×

× ×× ××××

××××

×××

×

× ×

××

× ×××

××

× ××

× ×× ×

××

×

×○ ○○ ○ ○○○ × ×○○○

○○

○ ○○○

○○

○ ○○

○○

○○○ ○○ ○

○○ ○

○○○

○○

○○

○○

×

××

××

× ××

××

×× ××××

× ×

××

×

×××

×

×× ×

×

××

××××

×××

Grey Zone50%

Grey Zone50%

Expanded space of the former Grey zone

Expanded space of the former Grey zone

Patent pended

Page 8: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

New approach to the “KY method”by one discriminant function

Highly overlappedPositive Region Negative Region

○ ×○

○○

○○

×

○○

○ ○

○○

○○

○○

○○

○○

○○

○ ○

○○

○ ○○

○○

○○

○○

○○

○○

×

× ×

××

××

××

×

×

×

××

×

×

××

× ×

×

×

××

×

×

×

×

×

×

×

×

×

×××

×

××

× ×

×

×

×

×

×

××

×

×

×

×

××

×

×

×

×

×

×

×

×

× ××

×

×

××

×

×××

×× ×

××

×

×

×

××

××

××

××

×

×

××

×

×

××

×

×

×

×

×

× ×

×

×

×

×××

×

×

××

××

×

××

×

×

××

d-pos d-neg

Page 9: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Discriminant Analysis Fitting

Two model KY KY Fitting with DA

Single model KY KY Fitting with no DA

Model free KY Model free KY Fitting

A series of KY methods

All methods were Patent Pended

Discriminant Analysis Fitting

Tailor-made Modelingfor DA

Tailor-made Modelingfor Fitting

*Always carry perfect classification *Always high coefficient of determination

*Always carry high prediction ratio

Page 10: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

○○

○○

○○

○○

○○

○○

○○

○○

○○

○○

○○

○○

○ ○○

○○○

○○○

○ ○

○○

□Starting large sample set

○○○

○ ○

○○

○○

○○

○○

○○ ○○○

○○

○○

○○

○○○

○○

○○

○ ○

○○

○○ ○○○

○○

○○

○○

○○○

○ ○

○ ○

○○ ○

○○○

○○

○○

○○ ○○○

○○

○○

○○

○○

○○

○○

○ ○○

○○ ○○○

Calcu

lated

Observed Observed

○○

○○

○○○

○○

○○

○○

○○

○○

○○

○○○

○○

○○

××

×

×

×

×

××

××

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×××

×

×××

××

×

×

×××

×

××

×

×

×

×

×

××

×

××

×××

×

×

××

×

×

×

×

×

×

×

×

××

××

××

××

×

××

××

×

× ×

×

×

×

×

×

×

××

×

×

××

×

×

×

×

××

×

×

×

×

×

×

×

××

×

×

×

×

×

× R=0.9

Inner samples(Gi)

Calcu

lated×

×

×

×

×

×

××

××

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

××

×

××

×

××

×

×

×××

×

××

×

×

×

×

×

×

×

×

×

×

×× ×

×

×

××

×

×

×

×

×

×

×

×

×

×××

××

×

×

×

××

××

×

× ×

×

×

×

×

×

×

××

×

×

××

×

×

×

×

××

×

×

×

×

×

×

×

××

×

×

×

×

×

×

Y(all)=β1x1+β2x2+・・・・・+βnxn+Const.

Calcu

lated

Observed

××

×

×

×

×

××

××

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

××

×

×××

×

×

×

×

×××

×××

×

××

×

×

×

××

×

×

×××

×

×

××

×

×

×

×

××

×

×

×

×

××

××

×

×

×

××

×

×

×

×

×

×

×

××

×

×

× ×

×

×

×

×

×

×

×

×

×

×

×

×

×

×

× ×

×

×

×

× ×

×

×

×

×

△△ △

△△

New sample space generated from

outer space samples

Calcu

lated

Observed

KY Fitting with DA

Patent pended

Page 11: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

◆KY method for fitting methods (Will be soon coming)Fish: 96 hours LC50、Number of samples: 791、Log(1/LC50_Mm) (Max/Min) : 6.376 / -2.963

◇ Data analysis by ordinal linear regressionStep1:Inner sample setNumber of samples:779, Number of used parameters:28, Confidance ratio:27.8R2:72.8, R:85.3, F-value:71.7, CV:69.6

Page 12: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Step1:Outer sample setNumber of samples:393, Used parameters:29, Confidance ratio:13.6, R2:64.7, R:80.4, F-value:22.9, CV:57.5

Step1:Inner sample setNumber of samples:398, Used parameters:22, Confidance ratio:18.1, R2:96.2, R:98.1, F-value:428, CV:94.4

◇Fitting KY method Step1 (Inner sample set) Patent pended

◇Fitting KY method Step1 (Outer sample set)

Page 13: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

O

N

O

O

Pharmacologicalactivity

Physicochemicalproperties

ADMEproperties

Toxicity

Drug properties and compound structure

There are no relations between any two properties.

All properties are fixed when the structure is determined.

All properties must be optimized for developing drugs.

Page 14: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Activity + ADME + Toxicity + Property

“ Integrated” concept

All drug properties shall be considered at the same time

“ Integrated” in silico screening & drug design

“Integrated” concept for drug development

Page 15: A NEW CLASSIFICATION METHOD SUITABLE FOR TOXICITY ...insilicodata.com/pdf lists/conference/EuroQSAR2010.pdf · TEST SIS Property Activity ADME Toxicity In Silico prediction Wet Experiment

Flow of the “Parallel & One Step” D.D.

“Parallel & One Step” D.D.

PhaseI/II/III

Confirmation

TESTSY

NTH

ESIS

Property

Activity

ADME

Toxicity

In Silico prediction Wet Experiment