Classic Case Studies

John MacIntyre
0191 515 3778
john.macintyre@sunderland.ac.uk

COM362 Knowledge Engineering 1

Classic Case Studies
 The Classics
 DENDRAL:
 determine molecular structure of an unknown
compound
 started in 1965

 MYCIN:
 medical diagnosis system
 started in 1972

COM362 Knowledge Engineering 2

Classic Case Studies
 DENDRAL
 Developed at Stanford University in 1965
 Possibly the first computer program EVER to rival
human experts in a specialized field
 Determine molecular structure of an unknown
compound
 Used a modified form of “generate and test”
methodology

COM362 Knowledge Engineering 3

Classic Case Studies
 The DENDRAL Problem
 Chemist is presented with an unknown chemical
compound
 Chemist must determine the molecular structure
 Therefore needs to find out which atoms are in the
structure
 Needs to know how the atoms are connected to
form molecules

COM362 Knowledge Engineering 4

Classic Case Studies
 The DENDRAL Problem
 Data from mass spectrometer
 Not straight-forward!
 Molecules can fragment in different ways
 need to make some predictions about how molecules are
LIKELY to break
 sub-components of the molecule may be found in many
different compounds
 chemists therefore determine compound sub-components,
and apply constraints that other sub-components must
satisfy
COM362 Knowledge Engineering 5
Classic Case Studies
 The DENDRAL Problem
 Not a trivial problem!
 Consider the formula: C6H13NO2
 There are 10,000 isomers of this compound!!
 Each permutation can be uniquely identified
 Could simply generate each of the10,000 permutations
in turn and test
 Very expensive in computing time!
 There would like to constrain the generation of
candidate permutations to save time

COM362 Knowledge Engineering 6

Classic Case Studies
 Constrained Generation
 CONGEN:
 DENDRAL program for constrained generation of
complete chemical structures
 Manipulates symbols representing atoms and molecules
 Uses a set of constraints on how atoms can be inter-
connected
 Chemist can specify and vary the initial constraints (eg
based on experimental evidence)

COM362 Knowledge Engineering 7

Classic Case Studies
 Specifying Constraints
 Defining “constraining structures”:
 specify “superatoms” that compound must contain
 typically in organic compounds, rings or chains of
carbon atoms linked to hydrogens
 Defining other constraints:
 open for the chemist to hypothesize
 eg “compound must contain a carbon ring of 6 carbon
atoms” etc….

COM362 Knowledge Engineering 8

Classic Case Studies
 Assessing Candidates
 CONGEN may produce hundreds or thousands of
candidate structures
 First pass at assessing the candidates:
 Use basic rules of mass spectrometry to test candidates
and remove most unlikely ones
 MSPRUNE: another DENDRAL program which does this
 MSRANK: ranks remaining structures according to how
their graphs match expected graphs for known compounds

COM362 Knowledge Engineering 9

Classic Case Studies
 Scoring Candidates
 Peaks (features) in the spectral graphs are
weighted to represent their importance
 Weighted scores are produced to give the rank
ordering for each candidate structure
 Essentially this is a “hypothesize-and-test”
strategy

COM362 Knowledge Engineering 10

Classic Case Studies
 Evaluating DENDRAL
 Available on the network of Stanford University,
California
 Used by hundreds of people around the world
every day
 Has been used to challenge long-published
chemical literature successfully
 The first stepping-stone between “traditional”
problem solving and modern expert systems

COM362 Knowledge Engineering 11

Classic Case Studies
 Features of DENDRAL
 Uses information from domain experts to help
limit the search space for candidate structures
 Uses an explicit representation of knowledge -
fragmentation rules
 No real inference mechanism - iterative passes
through the rules controlled by user

COM362 Knowledge Engineering 12

Classic Case Studies
 The Keys to Success?
 DENDRAL was successful because:
 It did not set out to replace the expert, only to assist the
expert
 The search technique is based on a proven model of
knowledge with known mathematical properties
 There is a language which can be used to represent the
structures easily and is well specified

COM362 Knowledge Engineering 13

Classic Case Studies
 MYCIN
 Developed at Stanford University in 1972
 Regarded as the first true “expert system”
 Assist physicians in the treatment of blood
infections
 Many revisions and extensions to MYCIN over
the years

COM362 Knowledge Engineering 14

Classic Case Studies
 The MYCIN Problem
 Physician wishes to specify an “antimicrobial
agent” - basically an antibiotic - to kill bacteria or
arrest their growth
 Some agents are poisonous!
 No agent is effective against all bacteria
 Most physicians are not expert in the field of
antibiotics

COM362 Knowledge Engineering 15

Classic Case Studies
 The Decision Process
 There are four questions in the process of deciding
on treatment:
 Does the patient have a significant infection?
 What are the organism(s) involved?
 What set of drugs might be appropriate to treat the
infection?
 What is the best choice of drug or combination of
drugs to treat the infection?

COM362 Knowledge Engineering 16

Classic Case Studies
 MYCIN Components
 KNOWLEDGE BASE:
 facts and knowledge about the domain
 DYNAMIC PATIENT DATABASE:
 information about a particular case
 CONSULTATION PROGRAM:
 asks questions, gives advice on a particular case
 EXPLANATION PROGRAM:
 answers questions and justifies advice
 KNOWLEDGE ACQUISITION PROGRAM:
 adds new rules and changes exisiting rules

COM362 Knowledge Engineering 17

Classic Case Studies
 Basic MYCIN Structure
Physician User

Consultation Program

Dynamic Static
Explanation Program
Patient Knowledge Base
Data

Knowledge Acquisition
Program

Infectious Disease Expert

COM362 Knowledge Engineering 18

Classic Case Studies
 The MYCIN Knowledge Base
 Where the rules are held
 Basic rule structure in MYCIN is:
if condition1 and….and conditionm hold
then draw conclusion1 and….and conditionn
 Rules written in the LISP programming language
 Rules can include certainty factors to help weight
the conclusions drawn

COM362 Knowledge Engineering 19

Classic Case Studies
 An Example Rule
IF:(1) The stain of the organism is Gram
negative, and
(2) The morphology of the organism is rod, and
(3) The aerobicity of the organism is aerobic

THEN:
There is strongly suggestive evidence (0.8) that
the class of the organism is Enterobacteriaceae

COM362 Knowledge Engineering 20

Classic Case Studies
 Calculating Certainty
 Rule certainties are regarded as probabilities
 Therefore must apply the rules of probability in
combining rules
 Multiplying probabilities which are less than
certain results in lower and lower certainty!
 Eg 0.8 x 0.6 = 0.48

COM362 Knowledge Engineering 21

Classic Case Studies
 Other Types of Knowledge
 Facts and definitions such as:
 lists of all organisms known to the system
 “knowledge tables” of clinical parameters and the
values they can take (eg morphology)
 classification system for clinical parameters and the
context in which they are applied (eg referring to
patient or organism)
 Much of MYCIN’s knowledge refers to 65 clinical
parameters

COM362 Knowledge Engineering 22

Classic Case Studies
 MYCIN’s Context Trees
 Used to organise case data
 Helps to visualise how information within the case
is related
 Easily extended and adapted as more clinical
evidence becomes available

COM362 Knowledge Engineering 23

Classic Case Studies
 Example Context Tree
PATIENT-1

CULTURE-1 CULTURE-2 CULTURE-3 OPERATION

ORGANISM-1 ORGANISM-2 ORGANISM-3

DRUG-1 DRUG-2

COM362 Knowledge Engineering 24

Classic Case Studies
 MYCIN Control Structure
 Uses a goal-based strategy to attempt to solve, in
the first instance, a TOP LEVEL GOAL RULE
 Establishes sub-goals required to satisfy the top
level goal
 Therefore establishes the concept of backward
chaining

COM362 Knowledge Engineering 25

Classic Case Studies
 Top Level Goal
IF:(1) There is an organism which requires therapy;
and
(2) consideration has been given to any other
organism requiring therapy
THEN:
compile a list of possible therapies, and
determine the best one in this list

COM362 Knowledge Engineering 26

Classic Case Studies
 MYCIN Subgoals
 Sub-goals are a generalised form of the top-level
goal
 Hence sub-goals consider the proposition that there
is a particular organism
 Exhaustive search on all relevant rules to test this
proposition (until or unless one succeeds with total
certainty)
 More like exhaustive search than backward
chaining
COM362 Knowledge Engineering 27
Classic Case Studies
 Selection of Therapy
 Done after the diagnostic phase is complete
 Two phases:
 Selection of a list of candidate drugs
 Choice of preferred drugs or combinations of drugs
from the list
 Therapy rules use information on:
 Sensitivity of organism to drug
 Contraindications on the drug

COM362 Knowledge Engineering 28

Classic Case Studies
 Example Recommendation
IF:The identity of the organism is Pseudomonas
THEN:
I recommend therapy from the following drugs:
1 - COLISTIN (0.98)
2 - POLYMYXIN (0.96)
3 - GENTAMICIN (0.96)
4 - CARBENICILLIN (0.65)
5 - SULFISOXAZOLE (0.64)

COM362 Knowledge Engineering 29

Classic Case Studies
 Evaluating MCYIN
 Many studies show that MYCIN’s
recommendations compare favourably with
experts for diseases like meningitis
 Study compared on real patients with expert and
non-expert physicians:
 MYCIN matched experts
 MYCIN was better than non-experts

COM362 Knowledge Engineering 30

Classic Case Studies
 MYCIN Limitations
 A research tool - never intended for practical
application
 Limited knowledge base - only covers a small
number of infectious diseases
 Needed more computing power than most
hospitals had at the time!
 Doctors reluctant to use it
 Poor interface

COM362 Knowledge Engineering 31

Classic Case Studies
 Conclusions
 DENDRAL was a ground-breaking program as it
showed that computers could match experts in a
specific domain
 DENDRAL was always intended as an “expert
assistant”
 MYCIN was the first “expert system” which
included an inference control structure
 MYCIN is limited for practical use

COM362 Knowledge Engineering 32

Classic Case Studies
 Further Reading
 Introduction to Expert Systems
 P. Jackson, Addison Wesley, 1990
 Expert Systems: Principles and Programming
 J. Giarratano, G. Riley, PWS Publishing, 1994
 Artificial Intelligence: Tools, Techniques and
Applications
 T. O’Shea, M. Eisenstadt, Open University, 1984

COM362 Knowledge Engineering 33

Classic Case Studies