a) In all cases, we plar close cooperation with the users in
all aspects of the problem. Although the hasic isolation
procedures are the problem of each investigator, his knowledge of
the available facilities and their limitations can be an important
aiioto sample preparation and analysis of the results. This is
particularly true for collaborators who are unfamiliar with the
techniques of HRMS {[e.g., sample size and resolving power
1ecessary tt» separate the mass doublets that can be realistically
axpetted in different contexts).

b) The needs of the user comnunity will be varied. Drs. Duffield
and Smith will, in collaboration with the users, determine the
kinds of MS experiments which will be most useful, considering
sample complexity, stability, quantity, and so forth. We wish to
utilize fully the existing ressurce and our proposed extensions,
hringing to bear on a problem any techrique which is appropriate
and can b2 provided. This will include the full scope of
available experimental techniques in MS (LRMS, HRMS, GC/LRMS,
SC/HRMS, matastable defocussing, ultra-high resolution mass
neasirements) and available computer programs (see below).

c) Many problems will be amenable to treatment by computer
programs which exist or which will be leveloped, for example,
structural isomer problems or HRMS interpretation on compounds in
3 well-understood class. We will take the responsibility for
utilizing these programs where appropriate to assist in structure
alacidation problems. We will instruct members of the community
in use of the programs when programs are used routinely hy
collaborators.

57
TIT. METHODS

“oleztular structure elucidatio1 entails the intelligent and
patient application of a large body of krowledge to each specific
problen. The importance and relative difficulty of the problem
impel us to seek the powerful assistance of computer programs to
help chemists in their analyses. It is unlikely that such
prograns will ever replace chemists, especially because computer
programs are readily written only to focus on rather narrow
aspects of problems. Nevertheless, our past research is
reasonably forwarded as a demonstration of the computer's ability
to assist in practical biomoletular characterization although this
w2s a spinoff from theoretically oriented research.

In order +) meet the major objectives of this proposal we will
foztus our attention primarily on structure elucidation of
biomedistally important compounds through MS and AT. However, many
of the computer programs can already use information from other
analytical techniques. Sno we want to be able to think of
structure 2lucidation in the context of an ensemble of analytical
capabilities.

A. Enhancing the Power of the Mass Spactrometry Resource

Wo have developed a siqnificant resource consisting of
instrunentation (the Varian MAT-711 and ancillary equipment) and
souputear programs for instrument 2valuation, data acquisition and
reduction. PRoutine reduction of high resolution mass spectra to
elamental compositions and ion abundances without human
intervention provides the capability for efficient handling of
lacge volumes of high resolution mass spectra (such as will result
From GO/HRMS runs). The development of the GC and of the GC/MS
sombinatior is in the excellent hands of Ms. Annemarie Wegmann,
who is responsible for operation of the complete system. We now
have more than two years of operational experience with the MS,
the $C and related equipment under a wide variety of experimental
tonditions.

Yona of the resource-related research discussed in this proposal
sar bo carried out without significant quantities of mass spectral
Jata. The existence and extensions of the MS resource, the
davelopment of computer techniques and the applications to
biomedical problems demand an efficient mechanism for acquisition
and reduction of MS data, and eventual transmission of the data to
the SUMPY cesource. Thus, operation of the MS requires
substantial computer support to deal with the large volumes of
Jata proliuced by the system at high data rates. We feel that a
properly configured system of hardware and software should
provide, at a minimum, the following capabilities:

1) Detailed evaluation of the condition and performance of the
“S prior +> recording data on valuable samples, with feedback to
the operator.

L

sonditioning, peak detection and peak analysis.

2) A ztoordinated system of hardware and software for signal

Tur
3) Pata reduction techniques based on a compntei (not theoretical)
noiel of the MS, including peak shapes, mass/time function, and
resolving power as a function of mass.

4) Poak profile analysis for nultiplet detection and resolution.

5) Conpuiter control of scan rates, clock rates and optimun
analog and digital filtering parameters.

3) Some on-line feedback, to the operator, assessing the
performance of the svstem 4uring an experiment.

7) The system must deal with frequent or repetitive HRMS
2ans, ceagairing the capability for rapid storage and analysis of
tard? volumes of data.

Pravious support of our research by the NIH and NASA has given us
a fiecn foundation of programs and experience. We have, up until
the termination of the ACME computer facility (July 31, 1973) ,
Jemonstrated capabilities 1-5 above. We were precluded from
picsuing capabilities 6 and 7 due to the configuration of the ACME

facility.

The demise of the ACME computer facility and the subsequent
inzorporation of the PL/ACME language into a new IBM 370/158
Facility under Stanford auspices has forced a reevaluation of the
means for pvroviiing HRMS laboratory computing support. We had
oreviously depended exclusively on ACME for data reduction
prozessing. The ACMF transition poses both technical and fiscal
Jactisions in that the real-time support capabilities of the new
facility will he different from ACME's and the fee for service
basis of the facility requires an explicit budget allocation for
its use. Previously we had received ACME computing support
without charge as part of the core research effort. Since we were
thus required to revise our computing plans, we have explored a
number of options for near-term as well as longer term solutions.

As outlined in the attached annual report, we have chosen an
interim approach (through the end of the carrent grant, 4/30/74)
which mininizes near-term costs, including hardware and software
zo1iversions as well as operating expenses. This approach entails
connecting the MAT-711 spectrometar to the 3707158 computer
through an IBM 1300 interface. It allows use of the existing
2L/AZME programs but will have real-time response limitations at
least as severe as ACME had {which is inadequate for either
SC/LPMS or SC/HRMS). Our existing computing budget provides for
only a very low level of instrument utilization in this mode.

For 2a lonq2r term solution these constraints are unacceptable.
surrent estimates are that continued use of the inadequate
1390-370/158 connection and PL/ACME intaractive programs under
full instrument productivity would cost up to £4,100 per month.
Threa alternatives have heen investigated for improving technical
necfarmance and reducing cost. This review has resulted in our
surrent proposal to augment the existing mini-computer system
(PPP-11720) with local storage and arithmetic capabilities. This
stand-alon2 system wonld not support real-time, on-line data
reduction but would allow routine data acquisition and instrument
performanca evaluation, followed by off-line data reduction.

IT
awe wee ewan een ewe ee eee ee eee wee eK BD EP eH OTe we ee eee

Alternatives considered include:
1) Modifisad 370/158 Connection

Ho discussed with personnel in the Stanford Center for Information
Processing (SCIP) various approaches for improving 379/158
Service. Detailed planning is still under wav within SCIP in
regari t> ceal-time support ani future pricing policies. Thus the
following ztonclusions are tentative. It appears that the
long-tern cost woul? be prohibitive to continue real-time data
acquisition by the 370/158. Insteai, 2 store-and-forward system
was proposed. This would entail an augmentation of the existing
PoP-11/29 front-end mini-computer with memory, disk, tape, anda
new interface to the 370/158, totaling about £28,000. This
approach is workable, if limited near real-time instrument
performance evaluations could be made to assur? satisfactory
instrument setup and data acquisition. It was recommended that
the existing software be converted fron PL/ACMF to a more
2fficient Language {such as FORTRAN) to reduce operating costs.
This would require approximately 4-6 man-months of effort. The
resulting decrease in operatiny costs could not be estimated in
tine for this proposal because the new SCIP pricing policies are
not formulated and inadequate 370/158 system analysis tools are
ypacational to avaluate our benchmarks in terms of detailed
cesourcte consumption. We have therefore budgeted an approach
based on the remaining two options with the understanding that we
will reconsider the SCTP option before proceeding with an
imolenentation should this proposal be funded.

2) SUMEX

fhe recently approved AIM-SUMEX PDP-10 facility will provide
gecessary computing support for the development and use of DENDRAL
AI programss. The ®S laboratory produces data which these
programs analyze and thus has a close relationship to the AI
research. The SUMEX computer could help in the off-line reduction
>f instrumant data, particnlarly during the early stages of the
proqact when the machine load will be relatively light (20-25%).
The present programs would require conversion from PL/ACME as in
aption (1), which would take 4-6 man-months. Such computing may
use From 15-30 minutes of CPU time per day, depending on the
amount of SC/MS work. While this approach saves operating
somputing costs, the front-end PDP-11/20 would require
zugmantation as in option (1) ($28,000) to allow store-and-forward
aperation with subsequent off-line data reduction on SUMEX. This
is needed because SUMFX is not configured to allow real-time
acquisition of the volume of data anticipated. This approach,
while the least costly, would entail a measurable use of the
PDOP-10 resources which we feel are better reserved for the
{itanded AIM-SUMEX applications. In addition, because of the
priorities anticipated for allocation of SUMEX to AI research,
particularly as loading increases, scheduling may he required
which will constrain the MS laboratory operation. For these
reasons, wa feel a better, even though slightly more expensive,
approach is a stand-alone PDP-11/29 data reduction systen.

3) Stand-Alone PNP-11/20

SY
The auamentations of the existing front-end PPP-11/20 required for
store-and-forward operation in conjunction with the 370/158 or
with SUMEY¥, come close to meeting the needs of a stand-alone data
system. In addition to the menory, disk and tape, an augmented
arithmetic capahility is needed to allow rapid floating point
talculations. A special device for this purpose costs about
£7,500. The SUMEX interface can be less sophisticated in this
case, however, accounting for the much lower data volume after
reduction, so that the total cost of the stand-alone system would
be $34,900, As with the other options, conversion of the present
programs wonld be required.

This approach, while slightly nore exvensive, has the advantages
af off-loaiing al] data logging and reduction functions from SUMEX
aad affords an adequate capacity for non-real-time, stand-alone
Jata reduction on the PDP-11/29. It furthermore allows more
freajon and responsiveness ir the operation of the MS laboratory
since aata collection or reduction can be scheduled without
worrying about the impact on AIM-SUMEX users. We therefore
aropose and have budgeted an augmentation of our existing
mini-computer system as a stand-alone data reduction facility.

ee ee a ne ee ee ee re ee ee ee ee ee ee ee ee ee ee ee ee ee

[The biomedical community (see ser Community, Sec. I.B.4 above)
Jasircing access to our facilities for structure elucidation have a
variety of problems, some of which can be solved hy existing
instrumentation and computer techniques, as noted above. However,
many problems consist of complex mixtures of compounds where
analysis by conventional GC/LRMS does not lead to unambiguous
solutions, and separation cof components on a preparative scale for
other spectroscopic analysis is difficult (9.g., see marine
sterdls, saction D, below). These problems are amenable to attack
by a system comprised of a GC/HR4S combination, the GC providing
s9paration, coupled with the MS operating at high resolution to
provide alamental compositions. Thas, upgrading of our current
system so that SC/HPMS data can be provided on a routine basis is
a desirabl2, and we believe necessary, st2p to solve mary of these
pcobloms.

Ha propose to continue the development of the SC/HRMS system while
maintaining existing capabilities of routine HRMS analysis and
3C/M3 where this efficiently responds to local needs. Many
nembers of the user community will require in addition to GC/HRMS,
HPMS analysis of relatively pure compounds or mixtures of small
nambers of compounds. we will provide this capability on an
interim hasis, using Stanford's TBY 370/158 system while the PDP
11/29 system is being upgraded.

We wera able, using the ACMF computer facility, to start
avaluating the operation of a GC/4S system at high mass
rasolutions. These experiments were hampered somewhat by the
Llinitations of the computer system used to acquire the data (only
aecasional, single scans were possible); they were necessarily
tiscontinued {as well as all HP#S operation!) upon the termination
of ACME. We do have, however, some benchmark figures for the
nerformance of the proposed system. Mixtures of fatty acid esters
(@.9., methyl palmitate and methyl stearate) gave good quality
mass maasurements (+-10 ppm) over a dynamic range of 100:1 for
sample sizes of the order of 0.5-1.0 micrograms/component during
190 sac/dAecade in mass scans {resolving powers 5,000-8,900).

We are haltingly continuing our evaluation of the SC/HRMS systen
aven without a data system, making measurements on individual ions
sf the mass standard and known materials in the GC effluent.

Thase data can be approximately translated into expectations
during dynamic scanning. We have performed an extensive series of
m-asurements on both methyl stearate and cholesterol (not
derivatized), the latter compound being more representative of our
~yrreait research problems. These measurements tend to confirm the
graliminaty data described above. Firmer data will be available
subsequent to the submission of this proposal.

We prorose to operate our existing GC/MS system under high
resolution conditions aiming toward optimization of resolving
powers, scan rates and GC and molecular separator operating
conditions to datermine the maximum usable sansitivity of the

SySstan.

#o recognize that the ultimate sensitivity will not approach that
attainable by photographic methods of recording; we feel that the
ability for on-line operation and evaluation of the operating
conditions of the MS partially offsets the sensitivity
jisadvantajes. We realize that some structure elucidation
problems will not be amenable to study because of the sensitivity
limitations: we feel, however, that many problems of interest to
the User Community can be studied effectively with this
performance capability. Rather than propose a research program to
inzrease the sensitivity of high resolution mass spectrometers
(e.g., MeLafferty, at.al., Anal. Chem., 44, 2282 (1972), dynamic
rascanning of peaks; Jet Propulsion Laboratory - chemical
multiplier emission/detector arrays, private communication to T.
Rindfleisch), we propose to identify our limitations and, with our
aollaborators, use discretion in selecting and preparing samples.

Further acrtalerations of technical capability to meet the state of
+ha art in sensitivity will require investments in hardware that
san he better justified at a later stage of a successful facility
progran . Meanwhile, other laboratories can be ©xpected to make
significant contributions to this important problem. Practical
rayacd for budget limitations is the main reason we do not press
this issue ourselves at the present time.

Significant improvements in sensitivity (with only small decreases
in mass measurement accuracy) can be achieved by operating the MS
at reduced resolving powers coupled with intelligent analysis of
thea resulting data to detect and resolve the potentially greater
aunber of xverlapping peak envelopes. This proposal is not
antirely new (e.g., see Smith, et.al., Anal. Chem., 43, 1796
(1971); Burlingame, et. al., in "Computers in Analytical
Chemistry, C.R. Orr and J.A. Norris, Fd., Progress in Analytical
thomistry, Vol. 4, Plenum Press, New York, N.Y., 1970, Chap. IIT).
We can, however, significantly extend these earlier techniques by
atilizatior of our multiplet rasolution algorithm. This algorithm
2mbolied in a computer program, has been shown to increase the
effective resolving power of the ™S up to a factor of three. It
hases its »0eration on a dynamic model of peak shape computed
Yirectly from the data. For computational efficiency and to avoid

56
spicious information, this algorithm would be best implemented as
a post-processor, basing its search for multiplets on the results
of prior elemental composition determination.

The ability to detect and analyze for unresolved peaks is mediated
by consideration of the mass measurement accuracy of an MS systen.
These systems are capable of determining peak positions {and thus
nasses) to a small fraction of ths peak width. The high accuracy
af such neasurements (#- 2-10 ppm) can, in fact, be utilized to
letect and "resolve" multiplets in instances where the unresolved
spazies ar2 known precisely (see Burlingame, et al., ref. above,
For C4 vs. 13¢ doublet detection and resolution).

For instantes where the heteroatom content of a molecule is known
yr where the possibilities are reduced severely by chemical,
Spectroscopic and mass measurement heuristics, there may be a
ranqe of possible overlapping ions resulting from fragmentation of
the moleculs. These potential overlaps may be computed and then
used (in combination with the known resolving power and mass
yeasuremant accuracy of the MS and the measured mass of the peak,
assuming it was comprised of only one type of ion) to direct the
nultiplet resolution program.

As an example, we have computel the possible mass doublets for

various ranges of compositions (Lederberg, et al., to be a
published). A sample table for Cc, N, 3 =<4 is appended {Table 1).

Inly 28 of the 364 possibilities are shown, namely those whose

nass difference {e}) <.95 mass units.

N€ these 28, 13 show e>.03 and would he fairly easy to resolve,
cretuiring 175009 resolution at MW=150.

At the othar extreme, 5 doublets show e<.01 {CNG vs. H4YO4; C2H20
vse N3: T2N2 vs H4¥03: C3N vs #293; and T4 vs H2NO2) which would
demand special treatment for resolution.

The 10 doublets for which .01 =< e =< .03 pose the interesting
thallenges for tradeoff of resolution vs. sensitivity in the
context of given problems. For example, if N is absent, the only
ambiguities are C3 vs. H402 {e = -.02) and C4 vs 03 {(e = .015).

“uch as we would wish always to have unambiguous empirical
formulas for all ions, HRMS remains a valuable tool despite these
limitations. As shown by these examples, even moderate resolution
reduces the number of candidates to a manageably small number of
alternativas. Contextual and intarval jJata (within the spectrum)
can o2 us2i to trim these further at two levels: (a) pooling of
aoaak statistics to sharper decision probabilities on the presence
af haternatoms -- the fraqments are subsets of the molecule and
(bh) the assemblage of candidat? solutions under each of the
alternative formulae. Manifestly, computer processing can sort
dyut branches of decision trees that would soon exhaust human

patience.

These heuristics are bnilt ints the DENDRAL programs {solutions
based on fraqmentation theory), but are also applicable to table
look-up approaches.

We (ref. 29,33), and others (9.9., H.-K, Wipf, et. al., J. Amer.
Than. Soc., 95, 3369 (1973)) have illustrated the importance of

57
netastable ion determinations in automated structure elucidation
based on MS data. Data on metastable ions must be judiciously
salacted because of the time and sample normally required to
perform the measurements. Our programs are now capable of precise
specification of those experiments necessary and sufficient to
distinguish among a set of canlidate structures. We seek more
afficient ways of acquiring these selected instrumental data.

This can ba accomplished with minimal cost by developing the
hardware and software necessary to perform (defocussed) metastable
scans and zalculate the data. Much of the hardware, except an
accurate sensor for accelerating voltaye, already exists. We have
had considerable experience in peak detection on the software
siie; the calculations to determine transitions are simple. [t is
assumed that the operator would manually adjust the instrument to
the i2sirel "daughter" mass prior to initiation of the scan of
netastable origins ("parents") of this daughter.

The retent availability of revarsei-geometry instruments has
provided naw methods of metastable defocussing (e.g., Beynon,
at.,al., Anal. Chem., 45 (12), 1023A (1973)). We have illustrated
the power of these techniques in mixture analysis (ref. 69). No
"normal" qeometry instrument is equipped to perform these
neasuremants to determine all the daughters of a given parent,
information which is frequently more useful than the converse.
This infornation can be obtained, in principle, hy synchronous
variation of two of the three fields (magnatic/ accelerating/s
electrostatic daflection) in a very accurate way. We would like
to explore this possihility because we feel that this technique,
if feasibl>, would represent a significant upgrading of the many
standard geometry, double-focussing instruments in existence.

B. Computer Assisted Structure Elucidation

AS mantion24d above, some existing programs can be used immediately
For structure elucidation problems using MS data. The programs
have been iascribed in detail elsewhere andi are mentioned in the
section on existing capabiliti2s (Sec. I.B.E, above). The
Planner's performance, for example, is excellent precisely in the
areas where MS, by itself, is capable of definitive structure
analysis. The general intellectual flexibility of the human
chemist is beyond the reach of plausible programs. On the other
hand, where the history of a sample is known, so as to restrict
the potential classes of compounds and for classes where the rules
af MS fraqnentation are well understood, the program's performance
natches that of trained mass spectroscopists, the program also
affers soma advantages in its exhaustive and rapid analysis of the
Jata. Many structure elucidation problems of the user community
Fit into tais category and existing resources can fulfill these

needs.

Whether man- or computer-implemented, MS cannot solve all
stcuctaure 2lucidation problems, however. In such cases, recourse
is to other spectroscopic techniques if sample size permits. As
Aascribed in the introductory section, diverse information is
ni2c2d together to achieve a solution. Interactive computer
programs can assist in seqments of this procedure, with the
advantages of exhaustive evaluation of the data and the molecular

structures suqgested by these data.

5k
Tn our own and in planned collaborative work, we will call upon
the extensive facilities of the chemistry department for
acyuisition of additional spectroscopic data. These services are
Fiiaanced by fees, paid from existing research grants of the user
coumunity. There are sufficient documented examples of structure
>Laicidatio.r problems to obviate the requirement for extensive use
af these additional facilities in jJevelopment of the programs. On
the other hand, the intensive pursuit of mechanized "intelligence"
in the domain of MS requires more than availability of public MS
qJata . It requires the collaboration of skilled chemists actively
engaged in practical MS research and, at the same time, committed
to the exploration of innovations in the application of AI to the
solution of the problems

As i121 the oast, we will develop the computer programs through
zlos2 collaboration among Drs. Duffield ani Smith (and other
nembers of their groups) and the program designers and
proqrammercs. For us, this means daily censultation for discussion
9f strategy, extensions to the program, ani solutions to new
problems. In particular, we propose to continue software
development (on the AIM-SUMEX facility) as folllows:

1) The rezeantly completed structure generating algorithm will be
the core of our efforts to assist in structure elucidation. The
structure yanerator can guarantee that the correct solution is
somewhere in the list of possibilities. Additional programs, such
as th2 Planner allow us to avoid exhaustive generation in
practice. Some parts of the cyclic structure generator program
have not b2en extensively tested yet, and these tests will be the
First task to completes.

2?) The structure elucidation task is strongly directed toward
rejection of whole categori2s (2.9g., compound classes) of
solutions as quickly as possible by using as mich knowledge about
the themical history or characteristics of a sample as is
available. Details of spectroscopic data then define the
nolesular framework more precisely. Each step in this procedure
represents the application of constraints on the sat of possible
solutions. Computational efficiency demands that these
constraints be applied early in the generation process when the
structure generator is utilized.

Jo have made some effort to examine the kinds of constraints used
by scientists engaged in structure elucidation. We have begun
Yesigning strategies so that these constraints can he brought to
bear on th2 structure generator. Some of these strategies involve
niior changes to the existing program; others require significant
axtensions of existing generating functions. One approach which
Spams particularly attractive to us is presently under
Jevelopment. This approach will utilize the existing structure
yenerator, with some modifications, to generate a dictionary of
cyclic skeletons up to those containing a maximum of twelve
tertiary vertices. The dictionary will be a complete, irredundant
list of ring systems which contain no multiple bonds and no
tut-edges {acyclic parts). This dictionary will be organized and
ceye) so that many constraints can be implemented easily. The
dictionary will allow exhaustive specification of ring systems
with loubl2e bonds and/or aromaticity. The rings themselves can he
Labelled with heteroatoms to generate heterocyclic ring systems,

57
ar with acyclic radicals to qenerate substituted ring systems.

The existence of the dictionary will lead to greater computational
afficiency as it needs to be generated only once, and specific
configurations of rings (numbers, sizes, fusions) can be pulled
immediately from the dictionary.

#2 propose to continue these investigations so that a reasonable
variaty of constraints can be recognized and utilized effectively
by a computer program. This rapresents the first step toward
increasing the chemical knowledge of a program which views
molecular structures and their manipulation as mathematical
entities and transforms.

3) Present, effective use of the structure generator or its
subroutines for special problems requires a detailed knowledge of
the program. We propose to develop an interface between users and
the progran to remove this requirement. The interface would
contain elements of structure input and display routines anda
simple language for application of constraints. Portions of these
alamants are available from other workers (e.g., Richard Feldman,
NTH) and we would draw on these sources whenever possible.

Hy We prooose that initial efforts will be directed toward a
system where the scientist examines his own data and inputs his
fiirdings (in terms of allowed and disalloved structural features)
to the program as constraints. The generator would then provide a
list of possible solutions to he evaluated, followed by iteration
on this procedure.

5) Many structure elucidation problems can be characterized as
assembly 92£ sub-structures inferred from spectroscopic data into
complet? molecular structures. Although there are two instances
in the literature describing programs with the capability to solve
this problem (see S. Sasaki, "Determination of Organic Structures
sy Physical Methods, Vol. 5," F.C. Nachod and J.J. Zucherman, Ed.,
Academic Press, New York and London, 1973, p. 285; M.E. Munk, C.5.
Sandano, R.L. McLean, and T.H. Haskel, J. Amer. Chem. Soc., 89,
4158 (1967)), we do not feel these approaches fulfill the
regquicements for generating complete lists of structures and
avoiting duplicate structures. We have some strategies to solve
this problem, thus extending the scope of the generator while
tying i+ more closely to the methods used hy chemists engaged in
structure alucidation. Our existing structure generator has this
capability; as long as the sub-structures are connected only by a
single tond, no new rings are formed.

6) Wo wish to implement general rontines for finding molecular
ions from spectroscopic data in order to improve the general power
of the Planning program. The current Planning program depends on
yaving som2 metastable ion information with HEMS data, together
with knowledge of the structural class with special rules for the
class. We will incorporate strategies suggested by Biemann (K.
Biamann andi W.J. McMurray, Tet. Lett., 647 (1965)) and McLafferty
(8. Venkataraghavan, F. W. McLafferty, and G. FE. Van Lear, Org.
Mass Spectrom., 2, 1 (1969)) for finding molecular ions, but also
give the program the flexibility to use class-specific information
whan available. The procedure will be to use these kinds of
iaformatioar within a general heuristic search paradigna.

7} The sertion on aims indicated some longer-term directions

60
which might be pursued. Of these, we feel that the incorporation
af threa-dimensional information into the program is perhaps most
inportant {e.g., representation of three dimensional information,
molecular nodelling including steric factors). Lederberg has
previously discussed ways (Ref. 1) in which three dimensional
information can be considered in the generation and representation
of molecular structures. More recently, the work of Wipke (J.
Amer. Chem. Soc, in press; personal communication) in connection
with computer assisted organic synthesis has provided important
results which we would attempt to utilize to avoid unnecessary
Juplication of effort. We plan to collaborate with Dr. G. Loew
(Stanford genetics Dept.) to utilize her available programs on
nolecular orbital methods to determine local minima for

conformations.

Another longer term goal which we feel is both interesting and
important is the use of an extended Predictor (which we have
previously described in the context of MS) to assist in
Jistirnquishing amona potential solntions to a structure
alucidation problem. We have recently carried out some extensions
to the existing Predictor by incorporating the ability to suggest
natastable defocussing experiments. Further extensions to include
knowledge about other spectroscopic techniques and the information
which can be elicited from these techniques are clearly feasible
an@ could be a powerful extension to our computer assistance
afforts.

[. Theory Formation

Jna inpoctant aim of this project is to improve the existing
theory formation capabilities and thus provide more assistance to
scientists investigating regularities within classes of compounds.
This is a theory formation task at a very pragmatic level. The MS
theory that the program attempts to find is of the same form as
the one practicing mass spectroscopists use for structure
alicidatioa. Thus, resulting pieces of theory are extensions to
both the scientists! theory and the computer's theory of the
Jiscipline. To improve this program we need to complete the
Plan-GZenerate-Test prodram that has been started (as described in
th2 appendad annual report) and tune it over many test cases. We
also wish to make the programs interactive and easy to use so that
they are more readily accessible. This can he done when the
programs are transferred to the AIM-SUMEX facility.

we plan to apply the theory formation program to two different
kinds of data: (a) the data collected in the interest of
anierstanding the mass spectronetry of a particular class of
zoupounds, as was done for estrogenic steroids, and (b)
sollections of diverse data that may provide some insight into
tore ceneral fraqmentation mechanisms. For example, we hope to
fiid general rules analogous to the alpha-cleavage rule or the
stability a€ aromatic rings.

The INTSOM program mentioned in Section (1) is the planning phase
yf the theory formation program. It currently runs in batch mode
an Stanforits 360/67 computer. We wish to add an interactive
monitor to INTSU™ to give an investigator the ability to set up
his awn conditions for interpreting the mass spectra and to
sontrol th2 type of summary he wishes to see. For example, if he

6 /
is interested in the allowable hydrogen transfers associated with
one specific process the program could be instructed to produce a
yery sp2ocific summary. Also, we wish to add an interactive
program for answering questions about the results. Por example,
an investigator should he able to find out easily how many
procasses involve cleavage of a specific bond and how strong their
resulting MS peaks are.

The INTSOM program is now used routinely by mass spectroscopists
at Stanford engaged in investigations of the mass spectrometric
fragmertation of various classes of organic compounds, primarily
steroiis. A manuscript is now in preparation (Ref. 54) describing
the fragmeitation of progesterone and related compounds. The
program was used extensively in this work. We are now beginning a
letailed examination of the fragmentation of steroids related to
the anjJrostane skeleton, particularly the biologically important
testosterones. We propose t9 continue to use the INTSUM program
in its present form and as it is improved in support of these
studies.

The qanerator of rules that we now have, RULEGEN, does a credible
job of explaining the regularities summarized by INTSUM. It has
found, for example, the well-known alpha-cleavage fragmentation
process ani beta cleavage followed by rearrangement in the low
casolnition data for fifteen aliphatic amines. The program will be
extended. in two important ways to increase its utility: (i) the
proaram needs to be able to work with an increased number of
Jescriptive predicates in the generation of rules, and (ii) it
needs to bh2 given a more flexible reprasentation of complex
fraqmentation mechanisms so that it can more easily find rules
iavolving nore than two bonds.

We will continue working with low resolution MS data of the
150-200 nonofunctional aliphatic compounds studied previously in
the coritext of the performance program. These compounds are
well-understood and thus provile a gooi test of the program's
effectiveness. In order to insure generality in the theory
Formation programs, we will alse test the system against the high
resolution mass spectra of the 68 astroganic steroids. Since they
ace also wall-understood, these compounds will show how well the
proqram can deal with complex ring systems, multifunctional
ztompoands, cleavages involving more than two bonds, and high
resolution data.

The existing programs are in good working order - within definite
Linits - 8s) we expect to apply them to new sets of data from the
“S laboratory as interest arises. For example, as the high and
low resolution MS from marine stersls are collected we expect to
use INTSUM and RULEGEN (at least) to assist in the interpretation
ani generalization of these data. Since these problems will
advance tha state of knowledge of MS, it is not correct to look on
than as test problems. However, in the past the programs
Jeaveloped most rapidly when they were applied to unsolved problems
af interest to our colleagues in the chemistry department.

Por Yavelopment of the interactive programs, we will rely heavily
on the criteria of acceptahility by Stanford users. The programs
themselves will he written in INTERLISP on the SUMEX computer.
Tnittially, we will provide interactive access to the control
daraneteors of the programs in order to allow users to tailor their
rans to thair immediate interests. Later we hope to expand these
to allow interrogation of the programs with respect to both
sontents of the results and the program's reasoning steps.

>. Applications to Binmedical Problems

Wa can immediately offer te the user community the Planner, for
analysis of HP/MS in terms of molecular structure. The program is
insensitiva to the source of the 4S data, and we foresee
significant use of the program for analysis of spectra of mixtures
without prior separation and spectra from the GC/HRMS facility
without adiitional programming effort. Examples of applications
ar2as are summarized below.

42 wish to exploit our existing capabilities of the analysis of
birdlogical mixtures without prior separation (ref. 33). This
approach will prove particularly useful in studies of mixtures
which aro difficult to separat2 and analyze by GC. Phytoecdysones
related to ecdysone, an insect molting hormone, present such a
problem. GC of these compounds is very difficult, although
high-pressure liquid cthromatography has recently been used to
aarry out separations. This class of compounds represents an
interesting and valuable test case for our combined MS and
somputer tachniques, particularly the specification and subsequent
acquisition of metastable defocussing data for precise linking of
parent and fraqment ions in the spectrum of a complex mixture
(c2fs. 28, 33). Model compouris, mixtures and current structure
alactidation problems are available (Nakanishi, Columbia; Takemoto,
Tohoku University, Sendai, Japan). Although most users cannot be
sompletely specific as to the natur2 of their future structure
2lacidation problems, we feel that several of these problems can
bo handled by soch an approach.

As the structure generator and its extensions are developed
Further, we foresee continuing use of an interactive version
applied to specific problems of the user community. As an
example, the work in collaboration with the GRC project will
involve studies of several classes of compounds extracted from
hunan hody fluids (e.g., aromatic and alivhatic acids, various
classes of bases, amino acids and carbohydrates) which contain
representatives varying by substitutions about a small number of
molecular skeletons. The generator can define all isomers which
must be considered as possible solutions.

For those problems which are amenable to attack by library search
procedures, e.g., screening of GC/LRMS runs of marine sterols to
weed out known compounds, we propose to use these procedures and
to investigate extensions to them. using a procedure related to
that described by McLafferty (K-S. Kwok, et al., J Amer. Chen.
Soc., 95, 4185 (1973), we seek to Jatermine from modified library
search techrigques the known structures which yield similiar
spectra. Utilizing the DENDRAL structural manipulation routines,
we would then seek to determine those related structures (whose
Spectra are not in the library) which are possible solutions. A
library, including Wiswesser Line Notation names, exists (F. W.
WcLafferty, private communication) and would be of some utility in
this work.

The MS facility in conjunction with our programs will ke used in
studies of the following natur?:

63
1) Prof. Djerassi - we plan use of the MS facilities and computer
proqrams in ongoing research connected with existing NIH-supported
studies on steroids and marine sterols and continued collaboration
with Prof. Adlercreutz on estrogen mixtures isolated from body
fluids. Further collaboration with Prof. Adlercreutz will be on
structural studies of new estrogen metabolites whose presence in
nixtuirces has been inferred through our previous collaborative
efforts.

Phe work on marine sterols presently utilizes SC/L&MS and
frequently laborious separation procedures to isolate individual
fractions for HRMS analysis. GC/HRMS will be a significant
assistance in this effort. We plan MS studies of known marine
sterols (utilizing INTSUM) to jierive fragmentation rules, which
then will he used in the Planner to aid structure elucidation of

naw compounds.

We also plan further work on extensions of MS theory in the
steroid field, initially focussed on additional hiomedically
iauportant classes of steroids related to the pregnane
{progesterones) and androstane (testosterones) skeletons. This
work is cucrently being carriel out by Dr. Smith in collaboration
with two visiting senior scientists (Dr. Roy Gritter, Dr. Geoff
dromayv) currently on sabbatical leave fellowships.

2?) ‘Chemistry Department Collaborators - as indicated by the
responses summarized in the letters of interest (Appendix A),
thara is significant interest in use of the MS facility by other
NIH-supported members of the chemistry department. All those
Yisted are familiar with the technique of MS as applied to
structure elucidation problems. Most have usel MS frequently,
darticularly Prof. Van Tamelen in his studies of the cyclization
af squalene and related studies in the terpenoid and steroid
fieli. The interests of these collaborators are generally in HRMS
and 3C/HREMS, with occasional nse of other capabilities of the
systam. The types of compourds studied by this group and an
iriication of the amount of use expected are summarized in the
letters of interest.

3) Zenetics Research Center (GRC): (Profs. J. Lederberg, H. Cann;
Dr. A. Duffield)

The body fluids analyzed by GC/LEMS
to aate include urine, blood, amniotic fluid and cerebrospinal
Fliia. ach body fluid is fractionated into the following
compound classes:

a) organic acids and neutral compounds

b) amino acids

c) carbohydrates
which after appropriate derivatization are analyzejl by
SCYLRMS/conputer system. A library of known LRMS will serve as
the primary means of identifying metabolites from their
axperimentally recorded LRMS.

n those instances where the LPMS is insufficient for metabolite
Jantification GC/HRMS data will be necessary to determine the
composition of all ions in its mass spectrum. These data will
jreatly enhance the prospects of identifying the metabolite in
question.

ve ad

GY
It is known {on past performance) that if a compound is present in
body fluids at the level of 1 microgram per GC peak then good
qguality HR/MS will be recorded (ion amplitude dynamic range of
12109, mass accuracy 9f +-Spom) using the Varian MAT 711 mass
spectrometer. If the GC peak of interest contains insufficient
material for a HRMS scan then preparative 3C conld be used to
concentrate that portion of the chromatograph effluent prior to
SC/HRMS.

Prior to the demise of the ACME computer system (July 31, 1973) we
Javeloped a GC/HRMS system and applied it to the analysis of
axtracts from body fluids. The following example represents
cesults obtained with this system during its development. The
example us3d was a routine analysis and was run to determine the
capability of the overall system during its development and not as
an unknown sample of extreme interest.

The total ion plot recorded during the Lifetime of the GC/HRMS
analysis of an amniotic fluid is reproduced as Figure 1. A
complete high resolution scan was recorded on each of the peaks
shown in Figure 1. Filing time of the time-shared ACME computer
systam did not allow the system to operate in a repetitive scan
nole. For the sake of brevity only the GC/HRMS scan (# 1594,
Fiqurce 2?) ctorresponding to glutamic acid N-TFA O-n-Butyl ester
derivative is produced. (The corresponding GC/LPMS scan is Figure
3). The scan time per decade of mass was 10.5 seconds, the
resolution 6,500 and the matching tolerance for the assignment of
ampirical somposition set to 4 mmu. The rasults show that the
system was capable of accurate mass measurement with a dynamic
rarge in ion amplitude of about 33:1 in this instance.

Tha cassation of computer support for the GC/HRMS system did not
allow a HEMS analysis to be maie which was crucial to the
ideatification of a metabolite present in a body fluid. Since
that time however, several instances have arisen where GC/HRMS
jata would hava been collected in an effort to identify
natabolites not previously seen.

The expected sample throughput in the GRC project with existing
personnel is expected to approach 5 to 7 body fluids per week {(15-
21 GIVL2MS fractions to be run in the Genetics Department per
weak). On average GC/HRMS would be required on 1 - 2 samples per
waek,

The research interests of the Medical School collaborators
relative t> the proposed #S resource are summarized in the letters
of interest (Appendix A). The MS services required by this
cormunity will include GC/LRMS (Forrest, Sera, Kalman for drug and
Arug metaholite identificatior, Rabinowitz and Wilkinson for
prostaglandin identification, Robin for identification of
oxidized/reduced rejox pairs, Hollister for Marihuana metabolites,
Rarchas, neurotransmitters, Fair, polyamines and the prostatic
antibacterial factor in urine); GC/HRMS (Trudell, drug metabolite
identification, Kvenvolden, structure of amino acids and related
compounds plus samples as required from interests described under
SC/LRMS). In those instances where the biological extract contains
insufficient material for a GC/HRMS scan preparative GC, using
axisting instrumentation within the chemistry department, can be

6s”
used to concentrate the material prior to the GC/HRMS analysis.
If the mat2rial of interest is obtained relatively pure by this
technique then HRMS analysis using direct sample insertion into
the ion source would be utilized.

Dn

rom

4s mantioned above, several of the computer programs have
immediate utility for assisting with structure elucidation
oroblems. For example, the Structure Generator program can answer
structural isomerism questions independently of mass spectrometry,
{o.q. , to provide lists of isomers in conjunction with isomer
interconversion problems such as carbonium ion rearrangements).
Because the program will be able to generate complete lists of
isomers with (or without) some specified structural features, a
researcher can have confidence that no possibilities have been
yverlooked. Some interest in the structure generator has been
expressed by representatives of the pharmacentical industry. The
qyanrerator could be used to suggest complete sets of structural
alternatives for possible synthesis, once a physiologically active
congener has been identified.

Tn more general terms, the structure generator can he
cichly suqqyestive of new, unexolored areas of synthetic
arqanic chemistry. for example, the generator has heen used
by a graduate student in chemistry, Mr. Jan Simek, to
jiantify the space of possible Diels-Alder condensation
products consisting of six atoms of any combination of
carbon, nitrogen, oxygen, and sulfur ina

six-nembered ring with one double bond. A literature

search through the Ring Index revealed that many of the
ring systens have never heen reported.

66
QV. SIGNTIPICANCE OF PROPOSED RESEARCH

Structure alucidation is an important and difficult problem for
birmedical scientists. Many of them lack the detailed chemical
hackground necessary to he efficient in this endeavor. Generally
speaking, they also lack the frequently complex and expensive
2quipment {(e@.q., high resolution mass spectrometers) to provide
spectroscopic data to assist them in solving problems of molecular
struztara. We plan to provide the chemical and analytical
expertise to facilitate the solution of their structural problens.
This research aims at providing more powerful techniques for
Jetormininy molecular structures than are now routinely available.
In particular, we have proposed {a) providing extended MS services
az a means of collecting powerful analytic data for scientists;

(b) develosing (and extending) sophisticated computer programs to
assist with the interpretation of the jata from mass spectrometry
and 2lsewh2re, {(c) developing (and extending) novel computer
programs to assist with formulation of the rules of
interpretation, and {d) applying these state of the art techniques
to problems of biomedical relevance. Our research group is thus
iedicated to a broad-based attack on the applications of structure
alucidation to biological and hiomedical problems.

The proposed research not only holds promise for significant
long-term advances, it can have immediate henefits as well. Many
nembars of the hiomedical community at Stanford have called upon
the MS laboratory for assistance in the past and will continue to
Jno so in the future. The proposed resogurce will provide the
conduit for a substantial increase in the utilization of MS within
the Stanford biomedical community, The ability of the proposed
resource to interpret the experimental data it generates (enhanced
hy the close proximity of the resource and hiomedical community)
Should rasult in a successful program 9f interdisciplinary
research.

4R¥S is an important source of data for these problems, and
SC/HRMS is still more important. Previous investment by the NIH

a the Varian MAT-711 HRMS system at Stanford can be utilized now
ni built zpon for the future. Continued operation of the GC/MS
ystam will give the Stanford community access to state-of-the-art
spactroscopic techniques and to professional mass spactroscopists
h> can help with ongoing problems.

The comput2r programs themselves constitute a unique resource for
assisting with the structure datermination. The previous NIH
Jrant supported development of the programs. Tn part, we are
requesting funds to exploit these programs.

One of the most siqnificant aspects of this work is its
interdisciplinary view of solving molecular structure problems by
inzelligently directed search of the space of chemical graph
structures. As a result of posing the structure determination
problem in this framework, we have been able to further the
knowledge about structure elucidation in at least three ways.
Ficst, soma of the knowledge used by analytical chemists has been
nade more precise for use in a computer program. Second,
codifying such knowledge for the computer has led to the discovery

67
»f new research areas to extend our existing knowledge of MS.
Several publications listed in the bibliography (Refs. 42 and
following) are reports of exactly this kind of research. Finally,
the computar's systematic search through the space of possible
structures gives the practicing scientist the confidence that no
structures were merely overlooked. The efficiency of the program
depends on the exclusions of many whole classes of compounds, but
the componter will have rejectel those classes using precise,
axplicitly stated criteria.

Our recent work on Finding MS interpretation rules (theory
formation) can provide additional unique capabilities for
assisting with the problem solving. We wish to continue this
research bacause it offers hope for a solution to the problem of
Furnisking real-world knowledg2 to computer programs ~-- in
particulac to the computer programs that assist with structure
alucidation. This is a pressing problem in current AI research.
High performance programs, of which DENDRAL is most often cited,
lerive their power from large stores of knowledge. Yet there are
no routine methods for infusing such systems with knowledge of the
task domaii. We believe our research in theory formation holds a
cay to the solution of this problem.

oy
V. FACILITIES & EQUIPMENT

fhe Stanford Mass Spectrometry Laboratory will provide 4S services
on the Varian MAT-711 mass spectromater coupled with a
4awlatt-Packard gas chromatograph (Model 7610A). As service
fastrunents for more rontine mass spectral analyses, the
laboratory has a MS-9 and CH-4Y mass spectrometers.

Nata reduction is currently provided on Stanford's IBM 370/158
sompater ia conjunction with a front-end PDP-11/20 data
acquisition computer. (The PPP-11/20 presently has only the
capability for buffering peak profile data between the mass
spectrometar and the IBM 370/158 computer at the Stanford Computer
Teanterc.) An alternative to buying time on the 370/158 is proposed
and discussed in the bndget justification.

Tha AT projrams will be run on the NIH-sponsored AIM-SUMEX
tonupater facility (a PDP-10 conputer with the TFENEX operating
systan, 192K words of memory, and adequate peripherals for our
puroos2s). Running these programs on SUMFY will incur no charge.

67
VI. BT BLT OGRAPHY
A. D&@ND2AL PUBLICATIONS

{1) J. Lederberg, "DFENDRAL-64 - A System for Computer
Tonstruction, Enumeration and Yotation of Organic Molecnles as
Tree Structures and Cyclic Graphs", (tachnical reports to NASA,
also available from the author and summarized in (12)).

{la) Part I. Notational algorithm for tree

structures (1964) CR.57029
(1b) Part II. Topology of cyclic graphs (1965) CR.68898
{1tc) Part III. Complete chemical graphs;

enbedding rings in trees (1969)

{2) J. Lederberg, "Computation of Molecular Formulas for Mass
Spectrometry", Holden-Day, Inc. (1964).

(3) J. Lederberg, "Topological Mapping of Organic Molecules”,
Proc. Nat. Acai. Sci., 53:1, January 1965, pp. 134-139.

(1) J. Lederberg, "Systematics of organic molecules, graph
topology and Hamilton circuits. A general outline of the DENDRAL
system." NASA CR-48R99 (1965)

(5) J. Leierberg, "Hamilton Circuits of Convex Trivalent
Polyhedra (up to 18 vertices), Am. Math. Monthly, May 1967.

(6) S. L. Sutherland, "“DENDRAL - A Computer Program for
Zeneratirg and Filtering Chemical Structures", Stanford Artificial
Tntelliqence Project Memo No. 49, February 1967.

{7) J. Lele2rberg and &. A. Feigenbaum, "Mechanization of
Taductive Inference in Organic Chemistry", in RB. Kleinmuntz (ed)
Formal Rapresentations for Human Judgmant, (Wiley, 1968) (also
Stanford Artificial Intelligence Project Memo No. 54, August
1967).

(8) J. Lederberg, "Online computation of molecular formulas from
mass number." NASA CR-94977 (1968)

(9) B. A. Feigenbaum and B. G. Buchanan, "Heuristic DENDRAL: A Program
foc Generating Explanatory Hypotheses in Organic Chemistry", in
Proceedings, Hawaii International Conference on System Sciences,

3. K. Kinariwala and F. F. Kuo (eds), University of Hawaii Press,
196R,

(10) B. G. Buchanan, G. L. Sutherland, and E. A. Feigenbaun,
"Heuristic BFNDRAL: A Program for Generating Explanatory
Hypotheses in Organic Chemistry". In Machine Intelligerce 4 (B.
Yeltzer and D. Michie, eds) Fdinburgh University Press (1969),
(also Stanford Artificial Intelligence Project Memo No. 62, July

{11) F. A. Feigenbaum, "Artificial Intelligence: Themes in the
Second Recade"™. Tn Final Supplement to Proceedings of the IFIP658
Internatioazal Congress, Edinburgh, August 1968 (also Stanford
Artificial Intelligence Project Mamo No. 67, August 1968).

70
(12) J. Lederherg, "Topology of Molecules", in The Mathematical

Sciences - A Collection of Essays, (ed.) Tommittee on Support of
Qesearch in the Mathematical Sciences (COSRIMS), National Academy
af Sciences - National Research Council, M.I.T. Press, (1969), pp.
37-51.

(13) G. Sutherland, “Heuristic DENDRAL: A Family of LISP
Programs", Stanford Artificial Intelligence Project Memo No. 80,
March 1969,

(14) 3. Lederberg, G. L. Sutherland, B. G. Buchanan, FE. A-
Faiganbaum, A. V. Rohertson, A. M. Duffield, and Cc. Djerassi,
"Anplications of Artificial Intelligence for Chemical Inference I.
The Yunber of Possible Organic Compounis: Acyclic Structures
Tontaining C, H, OQ and NN". Journal of the American Chemical
Sociaty, 91211 (May 21, 1969).

(15) A. ™. Duffield, A. V. Robertson, C. Djerassi, B. G.
3uchanan, 3. Le. Sutherland, FE. A. Feigenbaum, and J. Lederberg,
"Application of Artificial Intelligence for CThemical Inference II.
Interpretation of Low Resolution Mass Spectra of Ketones".

Journal of the American Chemical Society, 91:11 (May 21, 1969).

(16) R. G. Buchanan, G. L. Sutherland, E. A. Feigenbaum, “Toward
an Understanding cf Information Processes of Scientific Inference
in the Context of Organic Chemistry", in Machine Intelligence 5,
(8. Maltzar and DPD. Mickie, eds) Edinburgh University Press
11979), (also Stanford Artificial Intelligence Project Memo No.
99, September 1969).

(17) J. Lederberg, G. L. Sutherland, B. G. Buchanan, and E. A.
Feigenhaum, "A Heuristic Program for Solving a Scientific
Inference Problem: Summary of Motivation and Implementation", in
8, Banerji & M.D. Mesarovic {eis.) Theoretical Approaches to
Non-Numerical Problem Solving, New York: Springer-Verlag, 1970.
(Also, Stanford Artificial Intelligence Project Memo No. 104,
Yovember 1969.)

{18) C. W. Churchman and B. G. Buchanan, "On the Design of
Inductive Systems: Some Philosophical Problems". British Journal
for the Philosophy of Science, 20 (1969), op. 311-323.

(19) G. Schroll, A. M4. Duffield, c. Djerassi, 8B. G. Buchanan, G.
L. Sutherland, E. A. Feigenbaum, and J. Lederberg, "Application
of Artificial Intelligence for Chemical Inference TII. Aliphatic
Fthers Diagnosed by Their Low Resolution Mass Spectra and NMR
Data". Joirnal of the American Chemical Society, 91:26 (December

17, 1969).

(23) A. Bochs, A. M. Duffield, G. Schroll, C. Djerassi, A. 3B.
Delfino, B. G. Buchanan, G. L. Sutherland, ®. A. Feigenbaum, and
J. Lederberg, "Applications of Artificial Intelligence For
themical Inference. IV. Saturated Amines Diagnosed by Their Low
Pesotution Mass Spectra and Nuclear Magnetic Resonance Spectra",
Journal of the American Chemical Society, 92, 6831 (1970).

(21) Y.™M. Sheikh, A. Buchs, A.B. Delfino, G. Schroll, A.M.
Puffielda, Cc. Djerassi, B.G. Buchanan, G.L. Sutherland, E.A.
faigenbaum and J. Lederberg, "Applications of Artificial
Intelligence for Chemical Inference V. An Approach to the

7/
Tompnuter Ganeration of Cyclic Structures. Differentiation Between
All the Possible Isomeric Ketones of Composition C6H100", Organic
Yass Spectrometry, 4, 493 (1979).

(27) A. Buchs, A.B. Delfino, A.M. Duffield, C. Djerassi,
B.%. Buchaian, E.A. Feigenbaum and J. Lederberg, “Applications of
Artificial Intelligence for Chemical Inference VI. Approach to a
3eneral Method of Interpreting Low Resolution Mass Spectra with a
Tomouter", Helvetica Chemica Atta, 53, 1394 (1970).

(23) F.A. Feigenbaum, 28.G. Buchanan, and J. Lederberg, "On Generality
and Proklen Solving: A Case Study Using the DENDRAL Program". [In
“achine Intelligence 6 (B. Meltzer and D. Michie, eds.) Fdinburgh
Iniversity Press (1971). (Also Stanford Artificial Intelligence

%rodect Memo No. 131.)

(24) A. Bachs, A.B. Delfino, ©. Djerassi, A.M. Duffield, B.G. Buchanan,
®.A8,. Feigenbaum, J. Lederberg, G. Schroll, and G.L. Sntherland,
"Tha Application of Artificial Intelligence in the Interpretation
of Low- Resolution Mass Spectra", Advances in Mass Spectrometry,
5S, 3174.

(25) B.G. Buchanan and J. Lederberg, "The Heuristic DENDRAL Program
for Explaining fmpirical Pata". In proceedings of the IFIP
TSongreses 71, Ljubljana, Yugoslavia (1971). {Also Stanford
Artificial Intelligence Project Memo No. 141.)

(26) B.G. Buchanan, F.A. Feigenbaum, and J. Lederberg, "A Heuristic
Programming Study of Theory Formation in Science." In proceedings
xf the Second International Joint Conference on Artificial
Tntelliganze, Imperial College, London (September, 1971). {Also
Stanford Artificial Intelligence Project Memo No. 145.)

(27) Buchanan, B. G., Duffield, A.M., Robertson, A.V., “An Application
af Artificial Intelligence to the Interpretation of Mass Spectra",
Mass Spectrometry Techniques and Appliances, Fdited by G. W. A.
Milne, John Wiley & Sons, Inc., 1971, p. 121-77.

(28) D.H. Smith, B.G. Ruchanan, R.S. Engelmore, A.M. Duffield, A. Yeo,
7.A. Feigenhaum, J. Lederberg, and C. Djerassi, "Applications of
Artificial Intelligence for Chemical Inference VIII. An approach
to the Computer Interpretation of the High Resolution Mass Spectra
yf Complex Molecules. Structure Eluciitation of Estrogenic
Steroids", Journal of the American Chemical Society, 94, 5962-5971
(1972).

(29) B.S. Buchanan, F.A. Feigenbaum, and N.S. Sridharan, “Heuristic
Theory Formation: Data Interpretation and Rule Formation". In
Machine Intelligence 7, Edinburgh University Press (1972).

(30) Lederberg, J., "Rapid Calculation of “olecnlar Formulas from
Yass Values". Jnl. of Chemical Education, 49, 413 (1972).

(21) Brown, H., Masinter L., Hjelmeland, &., "Constructive Graph
Labeliag Using Double Cosets". Discrete Mathematics {in press).
(Also Computer Science Memo 318, 1972).

(32) 8B. G. Buchanan, Review of Hubert Dreyfus’ "What Computers Can't
No: A Critique of Artificial Reason", Computing Reviews (January,
1973). (Also Stanford Artificial Intelligence Project Memo No.

7a
Tat)

(33) D0. 4. Smith, B. G. Buchanan, R. Ss Engelmore, H. Aldercreutz and
2. Dierassi, “Applications of Artificial Intelligence for Chemical
Tafererce IX. Analysis of Mixtures Without Prior Separation as
Illustrated for Estrogens". Journal of the American Chemical
Society 95, 6078 {1973).

(34) DP. HL Smith, B. G. Buchanan, W. C. White, F. A. Feigenbaum,
™. Djerassi and J. Lederberg, "Applications of Artificial
Tnhtelligqenze for Chemical Inference ¥. Intsum. A Data
Tnterpretation Program as Applied to the Collected Mass Spectra of
Fstroqenic Steroids". Tetrahedron, 29, 3717 (1973).

(35) B. G. Buchanan and N. S. Sridharan, "Rul2 Pormation on
Non-~domogeneous Classes of Objects". In proceedings of the Third
Tnternational Joint Conference on Artificial Intelligence
(Staaford, California, August, 1973). {Also Stanford Artificial
Tntelligance Project Memo No. 215.)

(36) DP. Michie and 2.G. Buchanan, "Current Status of the Heuristic
DENDRAL Program for Applying Artificial Intelligence to the
Interpretation of Mass Spectra". August, 1973.

(37) #. Brown and L. Masinter, “An Algorithm for the Construction
of the Graphs of Organic Molecules", Discrete Mathematics (in
press). Also Stanford Computer Science Department Memo
ZTAN-CS-73-361, May, 1973)

(39) D.H. Smith, L.M. Masinter and N.S. Sridharan, "Heuristic
NENDRAL: Analysis of Molecular Structure," Proceedings of the
NATO/CNA Ailvanced Study Institute on Computer Representation and
Manipulation of Chemical Information, in press.

(39) RB. Carhart and Cc. Djerassi, "Applications of Artificial
Intelligense for Chemical Inference XI: The Analysis of C13 NMR
Nata for Structure Elucidation of Acyclic Amines", J. Chem. Soc.
(Parkin IT), 1753 (1973).

(40) L. Masinter, N. Sridharan, and D.H. Smith, "Applications
of Artificial Intelligence for Chemical Inference XTI: Exhaustive
Saneration of Cyclic and Acyclic Isomers.", suhmitted to Journal
of the American Chemical Society.

(41) L. Masinter, NS. Sridharan, 8. Carhart and D.H. Smith, "Applications
of Artificial Intelligence for Chemical Inference XTII: An
Algorithm for Labelling Chemical Graphs", submitted to Journal of
the American Chemical Society.

(4?) The Determination of Phenylalanine in Serum by Mass
Sragqmentography. Clinical Biochem., 6 (1973). By W.E. Pereira,
V.A. Bacon, Y. Hoyano, R. Summons and A.M. Duffield.

(43) The Simultaneous Quantitation of Ten Amino Acids in Soil
Fxtracts by Mass Fragmentography. Anal. Biochem., 55, 236 (1973).
Ry W.E. Pereira, ¥. Hoyano, %.F. Reynolds, 2.E. Summons and A.M.
Duffield.

(44) An Analysis of Twelve Amino Acids in Biological Fluids by Mass
Fragmentoqraphy. Anal. Chem., in press. By R.E. Summons, W.E.
Pareira, H.3. Reynolds, 7.C. Rindfleisch and A.M. Puffield.

(45) Tha Quantitation of B-Aminoisobutyric Acid in Urine by Mass
Fraqmentogcaphy. Clin. Chim. Acta, in press. By W.E. Pereira,
P.E. Summons, W.E. Reynolds, T.C. Rindfleisch and A.M. Duffield.

($6) The Determination of Fthanol in Rlood and Urine by Mass
Pragmentography. Clin. Chim. Acta, in press. Py W.F. Pereira,
R.E. Summons, T.C. Rindfleisch and A.M. Duffield.

Vie
2nblications Describing DENDRAL-~Related Research But Not Funded By
This Grant

(47) An Automated Gas Chromatographic Analysis of Phenylalanine in
Seram. Tlinical Biochem., 5, 166 (1972). By E. Steed, 4H.
Pereira, RB. Halpern, M. D. Solomon and A.M. Duffield.

(48) Pyrrolizidine Alkaloids. XIX. Structure of the Alkaloid
Frucifoline. Coll. Czech. Chen. Commun., 37, 4112 (1972). By P.
Sedmera, A. Klasek, A.M. Duffield and F. Santavy.

(49) chlocination Studies I. The Reaction of Aqueous Hypochlorous
Acid with cytosine. Biochem. Biophys. Res. Commun., 48, 880
(1972). Py W. Patton, V. Bacon, A.M. Duffield, R. Halpern, Y.
Yovyano, KH. Per2ira and J. Lederberg.

(52) A Stady of the Flectron Impact Fragmentation of Promazine
Sulphoxide and Promazine using Specifically Deuterated Analogues.
Austral. J. Chem., 26, 325 (1973). By M.D. Solomon, R. Summons,
@. Pareira and A.M. Duffield.

(51) Spectrometrie de Masse VITI. Elimination dtcan Induite par
Impazt Flectronique dans le
Tatrahyiro-1,2,3,4-Napthtal-ensa-diol-1,2. Org. Mass Spectre., 7,
357 (1973). By P. Perros, J.P. Morizur, J. Kossanyi and A.M.
Buffield.

(92) Chlorination Studies IT. The Reaction of Aqieous Hypochlorous
Acid with a-Amino Acids and Dipeptides. Biochim. at Biophys.
Reta, 313, 170 (1973). By W.E. Pereira, Y. Hoyano, R. Summons,
V.&. Bacon and A.M. Duffield.

(53) Spectrometrie de “asse. IX. Fragmentations Induites par
Tmpart Flecstronigue de Glycols- En Serie Tetraline. Bull. Chim.
Soc. France, 2105 (1973). By P. Perros, J.P. Morizur, J. Kossanyi
and A.M. Onffield.

(54) The Use of Mass Spectrometry for the Identification of
Metabolites of Phenothiazines. Proceedings of the Third
Trternational Symposium on Phenothiazines, Raven Press, New York,
1973, By A.M. Duffield.

(55) ChLlocination Studies IV. The Reaction of Aqueous Hypochlorous
Acid with Pyrimidine and Purine Bases. Biochem. Biophys. Res.
Tommun., 53, 1195 (19735. By Y. Hoyano, V. Bacon, RF. Summons,
W¥.F. Pereira, B. Halpern and A.M. Duffield.

(96) Mass Spectrometry in Structural and Stereochemical Problems.
CCXXYVIT. Flectron Impact Induced Hydrogen Losses and Migrations
in Some Aromatic Amides. Org. “ass Spectry., in press. By A.M.
Nnf field, GS. deMartino and C. Djerassi.

(57) Stable Isotope Mass Fragmentogqraphy: Quantitation and
tydroqen-Deuterium Exchange Studies of Fight Murchison Meteorite
Amino Acids. Geochem. et Cosmochim. Acta, submitted for
publication. By W.E. Pereira, B.F. Summons, T.C. Rindfleisch,

75