Extracting the Common Structure of Compounds to Induce Plant Immunity Activation using ILP 20/08/2015 Department of Industrial Administration, Faculty.

Extracting the Common Structure of Compounds to Induce Plant Immunity Activation using ILP

20/08/2015Department of Industrial Administration, Faculty of Science and Technology,Tokyo University of Science

Presenter: Atsushi Matsumoto

Introduction

Decreased production of agricultural crops

Using fungicides and pesticides

１

Pest Pathogen

Introduction

・ destruction of biota

・ health damage in humans

・ emergence of resistant pathogen

increasing the immunity of the plant

by plant immunity activator

２

Disadvantages of using fungicides and pesticides

Why is machine learning used?

・ the kind of compounds used in the experiments is enormous

→ the time required and the high cost of screening compounds

→ Efficiency of the screening by using machine learning

３

Objective

Extracting the Common Structure of Compounds to Induce Plant Immunity Activation

４

an example of Common Structure

Plant immunity

A defense system to protect plants from various enemies

５

Plant immunity

A drug that activates plant immunity

Plant immunity activator

Known plant-immunity activators

ProcessInput

６

compounds data 1.Structural features(e.g. atom , bond)2.Other features(e.g. Hydrophobicity)

Extraction

Transfo

rmati

on

1.Background knowledge(compound features expressed by the predicate)2.Positive or Negative

GKS(ILP system)

Learning

Output

Structure data

Predicate ７

Predicate List

Select several predicates as background knowledge

Predicate name(Argument) Detailatom(compound_name, atom_id, element) Type of atomsbond(compound_name, atom_id, atom_id, bondtype) Bonding state between atomsNum_AromaticRings(compound_name, num_AromaticRing) The number of aromatic ringsNum_Rings(compound_name, num_Ring) The number of ringsLogP98(compound_name, value) Lipid solubilityLogD(compound_name, value) Lipid solubility (consideration of the Ph)ring(compound_name, ring_id, atom_id, ringsize, ringtype) Type of ring structure

Dataset ８

positive negative total271 813 1084

High ROS compounds

Positive

Low ROS compounds

Negative

・ ROS(reactive oxygen species)If the ROS value is high, the compound is likely to be a plant-immunity activator

Experiment ９

①ILP Select several predicates as background knowledge

②SVM(To compare with ILP)

Using 77 features for learning

Approach tp fn tn fp Accuracy Precision Recall F valueSVM 123 148 703 110 0.762 0.528 0.454 0.488

Setting name tp fn tn fp Accuracy Precision Recall F valueILP1 92 179 699 114 0.73 0.447 0.339 0.386ILP2 116 155 644 169 0.701 0.407 0.428 0.417ILP3 127 144 605 208 0.675 0.379 0.469 0.419ILP4 88 183 712 101 0.738 0.466 0.325 0.383ILP5 131 140 572 241 0.649 0.352 0.483 0.407ILP6 139 132 568 245 0.652 0.362 0.513 0.424ILP7 165 106 523 290 0.635 0.363 0.609 0.455ILP8 165 106 542 271 0.652 0.378 0.609 0.467

Result １０

①ILP

②SVM(To compare with ILP)

Examples of the Common Structures１１

Rule number Interpretation Positive NegativeRule1 Atom C has a single bond with the aromatic ring. 27 10

There is an aromatic ring containing an atom S 20 8and atom C has a double bond with something.Two aromatic rings bond to each other 22 10and each aromatic ring have a single bond.An aromatic ring containing an atom N 15 3and An aromatic ring consisted of 5 atoms bond to each other

Rule5 An aromatic ring containing an atom S 14 2and another aromatic ring bond to each other

Rule2

Rule3

Rule4

Known plant-immunity activators

Rules

Rule2 Rule2 Rule1,2,5

Examples of the Common Structures１２

Rule1 Rule2

Rule3 Rule4 Rule5

＋

Conclusion １３

・ Extracting the Common Structure of Compounds to Induce Plant Immunity Activation

・ Although SVM F values slightly exceeded those of ILP, ILP tp values greatly exceeded those of SVM

・ Some rules were included in known compounds structure

・ As for other rules, there is a need for further investigation

Future work １４

・ In order to improve prediction accuracy, it is essential to improve background knowledge.(e.g. Add some functional groups to predicates)

・ In this study, it was a two-class classification. However, the ROS value is not a discrete value.It is necessary to change the threshold and the number of class.