To

FROM

DUBIECT.

Date: October 2, 1973

J. Lederberg
C. Djerassi
E. Feigenbaum

B. Buchanan, D. Smith,
T. Rindfleisch A. Duffield

DENDRAL Goals and Mike Oxman's Visit

The attached outline of long-term and short-term goals for

Parts A & C covers research items that we believe are feasible.

We do not know how well they fit with NIH goals. Therefore, if

appropriate, we would like to find a tactful way of getting Mike
Oxman's reactions to these items.

We are prepared to offer the available software as a service to
others (e.g., on SUMEX) in support of specific resource activities,
if that is desirable. Making one set of programs accessible to a
large community is preferable to exporting copies of those programs.

In short, we hope Mike's visit will help us arrive at a set of
guidelines for writing the DENDRAL renewal proposal.

 

 

k/
Att.
ERR
poy -
4) —

 

 

 
September 26, 2973
JUNDRAL Renewal: Potential Research Topics
PART A: Near-term

I) PLANNER Improvements.
a) Better mechanism for input, of rules for classes of compours.: «.
simple superatoms (e.g. -N- ).
b) Incorporate existing, extended molecular ion determication rou. .n>

TI) PLANNER Utilization
a) Choose areas in which program is competent. Compound class “ron
or severely restricted; mass spectral fragmentation rules
b) Suggest - Marine sterols; juvenile hormones; analysis o2 en! 2
bodily fluid components in support of shared resource Part B; otne"
classes of steroids.

 

ITI) Structure Generator

a) Develop an interactive version of complete Generator for ¢ .c...sts!
use.
b) Apply complete Generator to problems for which it is suitea, i.e.,

where necessary constraints may be supplied by interactive
guidance; e.g., isomer interconversion problems, labeliing src7b.--
ce) Explore in depth the problems of constraint implementation,
determine if existing algorithm is suitable, or new approeches c:¢
required.
d) Continue development of ancillary support for the Structure
Generator: extensions to CATALOG; enumeration (countinz) alg
for verification; carbocyclic ring index; PLUME catalog; non-
graph catalog.

4
F

a
ae

PART A: Long-term

The activities summarized in the following sections are directed tower.
development of software systems to attack the general problem of comsut>.-
aided or computer-directed molecular structure elucidation. The heuris is
search paradigm as embodied in the plan-generate-test strategy is not onl,
the most elegant approach, but is necessary for confidence in answers
(thoroughness) and, besides, we know how to do it.

I) PLAN

a) Chemists' inferences - an experienced chemist is capable o:
excellent structural inferences from diverse spectroscopic
Initial efforts would concentrate on chemists! planning ru_.u.
(constraints) coupled to an interactive generator.

b) Program inferences - develop and study the performance of srorrn..
designed, like the planner for mass spectrometry, to deveiox
structural inferences automatically from other types of spectrcscovl:
data.

c) Explore planning strategies to coordinate planning based cn dais
from different sources interpreted by different experts, one “%e
source.
DFNDRAL Renewal: Potential Research Topics
September 28, 1973

Page 2
II) GENERATE

a) Develop a structure generator which is knowledgable about chemistry;
i.e., a generator which is designed with the types of constraints
which must be applied in mind.

b) Develop a sophisticated chemists' interface to the generator
‘(presumably interactive) which allows input information in chemical
"language" independent of the inner workings of the generator.

c) Theta-DENDRAL - based on a particular problem, knowledge of
constraints and knowledge of the generator, develop a strategy for
solving the problem.

III) TEST
a) Extend existing Predictor to other spectroscopic techniques.
b) Develop the capability for examining lists of candidate solutions

c)

to determine how they differ.

Coupled with (b), suggest experiments for differentiation of
groups of structures.
DENDRAL Renewal: Potential Research Topics
Page 3

PART C: Near-term

I)

IT)

IIT)

INTSUM Extensions

a) Analyze summaries of mass spectra with respect to known mechanisms,
such as alpha-cleavage, in order to increase immediate utility.

b) Develop an interactive version of the theory formation program
that will answer specific questions about proposed new rules.

INTSUM Utilization

a) Interpret and summarize the mass spectra of important compounds
for which no theory exists.

b) Confirm existing theory before adding it to the performance
program,

Bond Environment Analysis
Analyze the bond sites of molecules to determine common features
influencing fragmentation.

PART C: Long-term

I)

IT)

Mass Spectrometry Theory

a) Develop capabilities for automatically modifying the performance
program's existing mass spectrometry theory.

b) Develop the capability of selecting the level of theory within
which rules best explain the data.

Extensions of Theory Formation Ideas

a) C1 NMR - develop a set of rules for interpreting C13 NMR deze
in much the same way as we have for mass spectrometry date.

b) Program Writing - extend the ideas to the domain of computer
programming. For example, find features of programs that cause
common problems or find a set of productions that '‘explain' 4
given set of input/output pairs.