Machine Learning network models from data Stephen Muggleton Syngenta university innovation Centre Imperial College, London September, 2012

Overview Systems Biology Machine Learning Biochemical network learning Automated experiment selection Conclusions

Systems Biology: The CISBIC Vision

Imperial College London

Machine learning

Logical

Probabilistic

Mixed

Decision trees

Neural nets

Bayes’ nets

Grammars

HMMs

SCFGs

Logic Programs

POMDPs

BLPs

Inductive Logic Programming Background knowledge. Protein sequence, partial grammar, incomplete biological network. Examples. Molecules, annotated sentences, temporal traces of up/down regulation. Hypothesis. Explanation of molecular 3-D shape, new clauses in a grammar, extra network annotation.

ILP for Systems Biology Robot Scientist

Metalog

CISBIC

Biology

Nature 2004

Nature 2006

Computing

Active learning

Networks

Dynamic

ETAI 2001

MLJ 2006

Modelling

Pathway database !A 
8
L  
.
-% 
% +!''

 %

 8 

-% 
!
 $ 

-% 
!
 $ 

-% 


"  








L  






:  
   )5$5,/.,.5,/@105,/.2050+'  6)5$5,/.2A5$,-%%+'  6!)5$5,/.2A5
B0'A'%'%1,B+'

!,/.20

!,/-0/

   6)5,/@105B05150C51(D )1    +    0'A'%'%1, %5.
  
% B+'

0'A'%'%1,

!$,-%%

6   ) 5  55,/@10+'

!$,-%% !,/@10

6)5$,-%%5B  
/(B+'

0'A'%'%-0 !$,0--

 6)5$5,/.,.5,,2/,+' 6)5$5,,2/,5B    
 B+'





!,/@%@

. .#  /          

+,+$

)%5%+' )%51+' ' '

!,"&


 6)#5#0+5


)#+5


)#57+5


 )85759+5


!,*, )95#0+5


'


'

+,+$ ' '

'/# %  )%5%50+'  6)%50+  )0515/+'  6)15/+ ' ' ' '

(  ) - 


! &   &# 
 

   ) 

Pathway abduction  

-.!/!"$!$"0  . )    )   )   1 )  ,2 )  (
)  * -."#0  ,)   )   )  ()  *)  +)  

 ,

3"#&
0  ,)  )  ()  *&


 3"#

"0$"#!45!$!/&!"

, 

  


 

  ()  *)  +   




 

!

  



  


    
.
-




































 



    
 
2   !
!

I

!"#!

&   ' !"#$%

     



"#

    !"#$%

/
 
%

%
A
 
4BJ
.
- 


;  


 
 0 



0 4



4

 0 

4





 

0 

H1: codes(geneX,rA) H2: codes(geneX,rB) H3: codes(geneX,rC) H4: codes(geneX,rD) H5: codes(geneX,rE)



0 

E1: A E2: B E3: C E4: D E5: E E6: AB E7: BC E8: BE E9: AD E10: ABC E11: ABCD E12: ABCE 0 1 1 1 1 0 1 1 0 0 0 0 1 0 1 1 1 0 0 0 1 0 0 0 1 1 0 1 1 1 0 1 1 0 0 0 1 1 1 0 1 1 1 1 0 1 0 1 1 1 1 1 0 1 1 0 1 1 1 0

!" "

!" !

"

!!

: "

!

"

;!

;&

$ "


;'

!'

% !

9

;) "


;'

"

!

"

!

;(

!

;( !

;!

;)

!

;&

Conclusions

• Integration of diverse background knowledge • ILP produces readable rules • Adbuction for gap-filling in networks • Abductive approach for learning food webs