October 27, 1977 7) AGE, an attempt to build: a software package of AI techniques and methods for problem solving and hypothesis formation; and its associated user-interface. The superb computing facilities of the NIH-supported SUMEX-AIM timesharing facility will be available at no charge to this project. The SUMEX-AIM facility, with Prof. Lederberg as principal investigator, is a national resource for the application of artificial intelligence techniques to problems in biology and medicine. Resources to be provided will include all CPU-time and storage required. Those involved at Stanford will be operating through hard-wired or dial-up equipment to the SUMEX PDP-10, while those at the University of New Mexico will access the system through either the ARPA network or TYMNET. The SUMEX-AIM facility is a powerful interactive computing system open to a national community. Interlisp and other high level languages are available and supported by a large system staff. Many convenient text editors for developing programs are provided. The TENEX operating system supports flexible file handling and sophisticated storage management for a highly interactive computing environment. October 27, 1977 Appendix I GLOSSARY AGE Stanford project ("Attempt to GEneralize") to build a general package of AI methods. AI Artificial Intelligence AI "toolbox" A set of AI concepts embodied in programs that are general enough to be used to construct problem solving programs in many different domains. alkaline phosphatase An enzyme that removes terminal phosphates from nucleic acids, and whose optimum pH is alkaline. AT-rich DNA DNA containing a high proportion of Adenine + Thymine base pairs. Because of the Watson-Crick base pairing rules, A/T=1 and G/c=l, but (A+T) /(G4C) ratios can can vary widely between DNA of different species and even different regions of the same DNA molecule. B. subtilis A common soil bacterium often used in genetic experiments. bacteriophage Viruses that multiply in bacteria. base sequence A string of nucleotides in a nucleic acid. bottom-up process A program that puts together inferences from data without the benefit of global expectations and goals. CONGEN Constrained Generator of molecular structures for DENDRAL, dalton A unit of mass equal to that of a single hydrogen atom. data-driven procedure bottom-up process demon a procedure in a_ program that is triggered by an event, aS opposed to being executed in the "normal" execution of a sequential program. 52 October 27, 1977 denaturation The loss of the native configuration of a macromolecule resulting, for example, from heat treatment, extreme pH changes, chemical treatment, or other denaturing agents. It is usually accompanied by the separation of strands (in DNA) and the loss of biological activity. DENDRAL Heuristic program for generating and testing organic molecular structures/as candidate explanations of empirical data. digestion With reference to enzymes, implies the cleaving of chemical bonds in the target molecule. For example, exonucleases "erode" or remove terminal nucleotides, restriction enzymes cut at the internal recognition sequences. dimer A concatenated DNA structure consisting of two identical constituents. discrimination experiment A series of experimental steps designed to conclude whether structures are identical or not. DNA Deoxyribonucleic acid. A polymer of deoxyribonucleotides (see nucleotide definition). Can exist as double or single strands. The genetic material of all cells and the central molecule in molecular genetics. domain-specific critic A procedure which applies specific EcoRI EDNA electron genetics knowledge to problem solving, as opposed to general problem-solving knowledge. A restriction enzyme isolated from a strain of Ee coli that cleaves DNA at site-specific regions along the molecule. Its recognition site is 5'- GAATTC-3', The DNA-structure-editor for MOLGEN. microscopy Abbreviated EM. A high-resolution technique for visualizing material that uses beams of electrons instead of light rays. Resolutions of about 10°-7 cm are posible with biological materials, electrophoresis An experimental technique used to separate, purify, and measure the molecular weight 53 October 27, 1977 of molecules having an electric charge in solution. endonuclease An enzyme that cuts DNA backbone chains internally. enzyme Protein molecule capable of catalyzing a specific chemical reaction. E. coli A common intestinal bacterium: the most intensively studied organism except for man. event-driven procedure demon exonuclease An enzyme that digests DNA from the ends of strands. experiment planning An activity characterized by the production of a sequence of experimental steps to achieve a goal. focus rules Focus of attention procedures. Items of knowledge that guide a program to the most relevant parts of the problem or the most useful subroutines. gaps An internal feature of double-stranded DNA which is aregion of unpaired nucleotides due to the excision of a string on one strand. HEARSAY AI program written at Carnegie-Mellon University to understand spoken English. Integrates inferences made by multiple experts. hierarchical planning AI techniques refined by Sacerdoti which uses a hierarchy of descriptions to plan an efficient problem solution procedure, inspector Domain-specific critic INTERLISP A powerful extension of the LISP programming language. KRL-0 A programming language (knowledge representation language) developed at Stanford and Xerox, Palo Alto Research Center. ligase ’ An enzyme capable of covalently joining parts of, or entire DNA molecules together. ligation The enzymatic joining together of DNA molecules. 54 linear DNA meta~rules October 27, 1977 Double stranded DNA that is not covalently closed at its termini. Rules for a program that mentions domain-specific rules, i.e, to prune or reorder the set of rules relevant for problem solving in specific contexts. molecular adapter A chemically synthesized segment of DNA MOLGEN monomer MYCIN nicks nuclease nucleotide pH phosphodiester that is utilized to join together DNA molecules which do not have complementary termini for ligation. Computer program for reasoning in molecular genetics. Main subject of this proposal. A single DNA molecule (or nucleotide) that has not undergone polymerization (viz. a unit character capable of assembly into a string). Medical diagnosis and therapy recommendation program developed at Stanford. A local interruption in the phosphodiester backbone of DNA. No genetic information is missing due to this structural anomaly. An enzyme which breaks chemical bonds in the DNA phosphodiester backbone. Consists of endonucleases and exonucleases. The building blocks of DNA consisting of a purine (Adenine or Guanine) or a pyrimidine (Thymine or Cytosine) linked to a deoxyribose sugar with a phosphate group also linked to adjacent sugar. Adjacent nucleotides are linked together through a phosphate group and a hydroxyl group on the sugar component (see phosphodiester). The negative logarithm of the effective hydrogen ion concentration or hydrogen ion activity in gram equivalents per liter. Used in expressing both acidity and alkalinity on a scale whose values run from 0 to 14; 7 representing neutrality, less than 7 increasing acidity, >7 increasing alkalinity. DNA exists in native form between pH values of 5 and 12, . The chemical link between adjacent nucleotides. The following diagram of its structure was drawn using CONGEN [6]: plan schema October 27, 1977 . / . Oo oO \ \/ 0 O P \/ /\\ P o 8600 //\ / Oo O- CH2 \ c-C / | CH \ Cc-C / BASE A sketch of a procedure describing the plan for an experiment in abstract, general terms. planning islands Partial solutions toa problem found bya planning rules plasmid poly-A region poly-c poly-G poly-T planning program. "Stepping stones" to a complete solution. Procedures or items of knowledge that aida program in constructing a problem solving plan. Extrachromosomal DNA molecules which are double stranded, circular, and supercoiled. They range in size from about 5*10°6 daltons to near 10°78. Small plasmids can exist in many (more than 50) copies per cell while large ones are maintained at one or two. They are often used as vectors for amplifying and transferring DNA from one organism to another. (sequence) A homopolymeric sequence of adenine nucleotides. Implies a poly-T region on the complementary strand. Homopolymeric cytidine nucleotides. Homopolymeric guanine mucleotides. Homopolymeric thymine nucleoties. 56 October 27, 1977 polymerase . Enzymes that are catalysts for nucleic acid chain growth, pre-conditions Premise clauses of conditional sentences that must be satisfied before the consequent actions are taken. production rules Conditional sentences used to encode inferential knowledge for a program. prototype . The type of unit created for representing information about general concepts. Features are defined by slots associated wih the prototype. restriction enzyme Site-specific endonucleases used frequently in molecular genetic manipulations. Allow previously impossible experiments to be performed due to their ability to cleave DNA at reproducible locations allowing rearrangements within and between molecules. RNA Ribonucleic acid. Typically single stranded, is a polymer of ribonucleotides connected by phosphodiester bonds. schema/rule schema/program schema An abstract, generalized representation of a concept or program. In MOLGEN, program schemata (or rule schemata) are represented as Units with slots defined for important features. SECS Chemical synthesis plarning program developed by . Prof, Todd Wipke (U.C. Santa Cruz). self-circularization Ligation of the ends of the same DNA resulting in a circular, covalently closed molecule. self-ligation Ligation of a DNA molecule to itself, resulting in in a circular molecule. Catalyzed by ligase. sequencing experiment A technique to determine the order of nucleotides in a strand of. DNA, slots Pre-defined features of objects for which values are sought. SMALLTALK Display~oriented programming language developed at Xerox, Palo Alto Research Center. | 57 sticky ends STRIPS SUMEX-AIM Teiresias TENE Thy gene October 27, 1977 A condition of partial single-strandedness at the termini of DNA molecules, allowing base pairing in that region. Restriction enzymes often leave sticky ends, greatly facilitating the rearrangements of DNA, Robot plnning program developed at SRI. NIH-sponsored computer resource for applications of artificial intelligence in medicine. AI program that acquires inference rules for MYCIN and guides MYCIN reasoning. Operating system for the DEC KI-10 system running at the SUMEX-AIM facility. A gene coding for the enzyme, thymidylate synthetase. This enzyme is crucial in enabling a bacterium possessing it to produce thymidine, a constituent of DNA, top-down process A program that works from general principles, Units vector world states 3'~end/ 5'-end testing data against expectations and goals, often working by dividing complex problems into simpler ones. Basic element of representation in MOLGEN. Units are organized in a hierarchy to facilitate the representation of class-subclass and prototype- instance relationships. Units are used for representing processes as well as concepts. A self-propagating DNA molecule that can be used to link DNA sequences of interest. Vectors can be one of several replicating plasmid or bacteriophage DNAs. Representation of the state of an experiment at any given time. The “world" for the program is the limited set of objects and operations relevant to a specific experiment. Related to the direction of the phosphodiester bonds in the backbone of DNA molecules. Each Strand thus has one 3' end and one 5' end. 58 October 27, 1977 Appendix IT EDNA -- The Editor for DNA The following is an actual session with the MOLGEN knowledge acquisition system recorded from SUMEX. Comments preceded by a semicolon have been inserted to clarify some aspects of the dialog. @ue :UE is the name of the Unit ;Editor. Here it is being called (Version 4-OCT-77 08:56:16) sfrom TENEX Welcome to the MOLGEN Unit Editor. Type ? anytime for assistance. The symbol : indicates that the editor is waiting for your input. Two characters are enough for command recognition. You may type ahead responses for a command. Name of Network: jerry ;Jerry is the name of an yexisting Knowledge Base on :file. :create testl root specialization 7A new unit TESTI is created. Give a value for the DESCR slot ;UE asks for documentation, Text Editor te: Test unit to demonstrate the DNA structure editor te: done :User indicates he is done swith documentation. Do you want to see what slots have already been filled? yes DESCR: (U) from ROOT Test unit to demonstrate the DNA structure editor MODIFIER: (U) from ROOT STEFIK CREATOR: (U) from ROOT STEFIK MODIFIED: (U) from ROOT 6-OCT-77 10:22:03 CREATED: (U) from ROOT 6-OCT-77 10:22:03 *Note that the system has rautomatically recorded the zauthor, date and time of the snew unit. You can now create new slots or edit old ones. When through type DONE EDIT: create substrate Datatype: dna ;DNA is a datatype. Role: r :"Role” controls transmission sof the value in the ssubstrate slot if we make 59 Is it a dynamic slot? n DNA Editor Copy or Create anew? create Segment Type: ? October 27, 1977 pspecializations of TESTI. ;Since the datatype is DNA, ;we get the DNA editor. 7A "2" may always be typed to tell the system to clarify swhat it expects for a ;response. Choose one of the following Segment Types. Type Description LE Length Segment. Indicates number of nucleotides in a region. BA Base Segment. Indicates actual Base Sequence. SI Cut Site for enzyme. Segment Type: length Length: ? sAnother "?"™ Indicate number of nucleotides as in the following examples: You Type Meaning 5 5 nucleotides 100 200 1K 1.3K R 1K 1.3K Length: 2.5k edna: print DNA Printer (Version 19-SEP-77 ) 1L (2500) edna: insert 1 3' bases attacg edna: print 1 234567 (2500) AT TACG edna: mirror 4 to 7 edna: print 1 234 5 (2500) ATT A AN Q~ 60 Between 100 and 200 nucleotides. Between 1000 and 1300 nucleotides. (K=1000) No Spaces! Same aS above except RNA instead of DNA. , ;The initial segment is now ;Sspecified. We may now issue yany legal EDNA command. ;EDNA presents its structures ;pictorially. Segments are rreferenced by number. :"Mirror" means to add a sparallel DNA strand. 7EDNA knows about (Renumbering 11) October 27, 1977 ;complementary bases. sBreak a bond. A Break ;command can specify 5', 3', 7or H bonds. :Segment 11 is depressed sto indicate the broken bond. 7A similar notation is used ;to indicate Hairpin loops. ;EDNA can "undo" any of its structure changing commands ;User is finished editing sthis structure. He returns to sthe slot editor. STEFIK STEFIK 6-OCT-77 10:22:09 6~-OCT-77 10:22:03 A T GC 11109 8 edna: break 4h edna: print 1 2345 67 (2500) ATTA CG / —— ~-T GC A 1098 il edna: connect 7 3' 8 edna: print 1 2345 67 (2500) ATTA C G8 | /---~--| —--T GC A 109 8 11 edna: undo CONNECT undone. edna: done EDIT: print all DESCR: (U) from ROOT Test unit to demonstrate the DNA structure editor MODIFIER: (U) from ROOT CREATOR: (U) from ROOT MODIFIED: (U) from ROOT CREATED: (U) from ROOT SUBSTRATE: (R) *Top* 1 2345 67 (2500) ATTA CG \— ——~7T GC A 1098 61 EDIT: done :done Save JERRY? no Bye 62 October 27, 1977 ;User is done with this unit. ;User is done with this sknowledge base. 7He doesn't save his changes rbecause this was just a ;demonstration. October 27, 1977 Appendix III A Genetic Planning Example This section is intended to extend the range of genetic examples for which MOLGEN is envisioned being applied. In particular, the recent cloning of the rat insulin gene in E. coli [38] has been achieved using a simple additional step to the usual experimental protocol, It is asserted that the genesis of this efficiency—improving step can be found in the relatively simple application of knowledge about enzyme properties and DNA ligation kinetics. The basic experimental outline is seen below, modified from [38]. It closely follows the 'classical' recombinant DNA methodology, with a few additional steps. The one we wish to focus on most closely in this discussion is the application of Alkaline Phosphatase to the plasmid vector after cutting of the plasmid by the restriction enzyme HindIII, ---~-----(T)y 5! : | Alkaline Digestion | - 3! | w----——== (T)y 5! | Reverse transcriptase 63 { Sl Nuclease Tomer (AZ 38 pwn manenn n= (T)Z 5! | | T4 DNA Ligase | + | 5' CCAAGCTTGG 3! | 3' GGPICGAACC 5! | October 27, 1977 plasmid HindIII site --//-- / -/f--- \ II Hl 1] {| HI | || L| \ / | HindIIt | restriction . | enzyme 5' CCAAGCTTGG---~~--—- (A) zCCAAGCTTGG 3! 5' paAGcTT-----— A 3! 3' GGTICGAACC-—--~—--—- (T) zGGTTCGAACC 5! 3" A-----TICGAp 5' | | | HindIII restriction enzyme | ALKALINE | | PHOSPHATASE 5' pAGCTTGG--------- (A) 2CCA 3! 5' ohAGCTT-—--A 3! 3’ ACC--------- (T) zGGTICGAp 5! 3' A-—---TICGAoh 5!' | | \ | \ / \ / \ 7s \/ | T4 DNA Ligase | * ~-AAGCTTGG-------—= (A) zCCAAGCTT—— / --TICGAACC—--~----- (A) zGGPICGAA-- \ || 7 cDNA l | {| I [| || li plasmid DNA lI \ / 64 October 27, 1977 The following steps were carried out. First, insulin messenger RNA was purified from B cells in the rat pancreas. This was reverse transcribed into a hybrid DNA/RNA structure by the use of avian myeloblastosis virus (AMV) reverse transcriptase and the RNA selectively degraded by raising the pH. A double-stranded DNA form was synthesized by incubating this with deoxynuclecside triphosphates and the AMV reverse transcriptase (a DNA polymerase could have been used). The hairpin at the end of the molecules and any non~base paired regions were removed with the single-strand specific nuclease Sl. The resulting molecular structure is termed. cDNA, or copy DNA, because it should contain the precise genetic information contained in the gene coding for the insulin messenger RNA. ‘This is the in-vitro synthesized segment that is to be cloned in bacterial recipients for amplification and analysis. A recently developed technigue for ligating chemically synthesized restriction site linkers (adapters) [35] to cDNA was used in order to produce cDNA molecules with cohesive termini after digestion with a restriction endonuclease enzyme. Ligating the resulting cDNA to plasmid DNA cut with the identical restriction enzyme would create a recombinant plasmid which could then be cloned ina suitable bacterial host. Specifically, a decamer linker containing a Site for HindIII was covalently joined to the ends of the cDNA with T4 DNA ligase, and then cleaved with HindIII; pMB9, a 3.5 million dalton plasmid conferring tetracycline resistance with a single site for HindIII, was also cut with the same endonuclease, Tne usual procedure would he to now straightforwardly ligate these two molecules together creating the desired recombinant molecule. Kinetic theory [11] suggests that in order to insure ligation of most of the cDNA molecules to plasmid DNA, it is necessary to add a molar excess Of plasmid DNA. However this would result in the majority of the plasmids simply self-circularizing without an insert of cDNA, and thus most the transformed cells would contain only pMB9 and not the desired recombinant plasmids. Here is where the novel step of removing terminal phosphates on the plasmid was generated. Several sources of knowledge need be brought to bear in order to understand the basis. of this new optimization step. Firstly, we need to know that alkaline phosphatase removes 5' terminal phosphates from the HindIII endonuclease-generated ends of the plasmid. Secondly, knowledge about the requirenent of the T4 ligase for a phosphate end configuration allows us to infer that rewoving the phosphate ends prevents self-Ligation of the plasmid DNA, Thirdly, the kinetic theory of ligation [1]] combined with a rule that says, "In a process that involves two or more competing components, you can optimize one process by inhibiting the other (s)", 65 October 27, 1977 should tell us that circle formation is now dependent on the insertion of a DNA fragment containing 5'-phosphorylated termini: the cDNA, Finally, since transformation is directly linked to the DNA source, the one step inference is: "Only recombinant plasmids will transform the recipient bacteria”. A side effect that needs to be dealt with is the fact that the recombinant plasmids generated after phosphatase and ligase treatments will have two nicks, represented as asterisks in the figure, and that this has no known effect on transformation efficiency. The application of alkaline phosphatese to remove terminal phosphates from a restriction enzyme-cleaved vector (e.g. plasmid) to eliminate self-ligation is a novelty that “should" have been obvious to any investigator working in this field. In fact, three or four years passed before Ullrich et. al. utilized it. One can only speculate as to the reasons why. However two related responses arise in this context, First, there are a very large number of DNA reagents available to the investigator (enzymes, chemical and separative technigues) so the number of possible combinations are vast. Secondly, especially with a well focused goal such as the Jigation optimization step discussed above, people tend to think along relatively stereotyped paths, e.g. previously ceveloped protocols. A computer system, such as MOLGEN, with a complex knowledge base and a good? set of heuristic rules, will be likely to uncover novel applications of well known tools, precisely along the lines of the example just presented. 66 October 27, 1977 should tell us that circle formation is now dependent on the insertion of a DNA fragment containing 5'-phosphorylated termini: the cDNA, Finally, since transformation is directly linked to the DNA source, the one step inference is: "Only recombinant plasmids will transform the recipient bacteria". A side effect that needs to be dealt with is the fact that the recombinant plasmids generated after phosphatase and ligase treatments will have two nicks, represented as asterisks in the figure, and that this has no known effect on transformation efficiency. The application of alkaline phosphatesé to remove terminal phosphates from a_ restriction enzyme~-cleaved vector (e.g. plasmid) to eliminate self-ligation is a novelty that "should" have been obvious to any investigator working in this field. In fact, three or four years passed before Ullrich et. al. utilized it. One can only speculate as to the reasons why. However two related responses arise in this context. First, there are a very large number of DNA reagents available to the investigator (enzymes, chemical and separative technigues) so the number of possible combinations are vast. Secondly, especially with a well focused goal such as the ligation optimization step discussed above, people tend to think along relatively stereotyped paths, e.g. previously developed protocols. A computer system, such as MOLGEN, with a complex knowledge base and a good set of heuristic rules, will be likely to uncover novel applications of well know tools, precisely along the lines of the example just presented. 66 October 27, 1977 References Bobrow D.G. and Winograd T., Experience with KRL-0: One Cycle of a Knowledge Representation Language, 5IJCAI:213-222 (1977) Bobrow D.G. and Winograd T., An Overview of KRL, a Knowledge Representation Language, Cognitive Science 1 (1977) Brachman R.J., What's in a Concept: Structural Foundations for Semantic Networks, BBN Report No. 3433 (1976) Brown R., Use of Analogy To Achieve New Expertise, Artificial Intelligence Laboratory, MIT, AI-TR-403 (1977) Buchanan, B.G. and Dennis Smith, "Computer Assisted Chemical Reasoning", in Proceedings of the III International Conference on Computers in Chemical Research, Education and Technology, Plenum Publishing, pp. 388-408 (1977) Carhart, R.E., Smith, D., Brown, H., and Djerallis, C., Applications of artificial intelligence for chemical inference. XVII. An approach to computer-assisted elucidation of molecular structure J. Am. Chem. Soc. 97:5755-5762 (1975). Davis R., Applications of Meta Level Knowledge to the Construction, Maintenance and Use of Large Knowledge Bases, Stanford Computer Science Department Report No. STAN-CS~76-552 (1977) Davis R., Generalized Procedure Calling and Content-Directed Invocation, SIGPLAN Notices, 12:45-54 (1977) Dershowitz N. and Manna Z., The Evolution of Programs: A System for Automatic Program Modification, Stanford University Artificial Intelligence Laboratory (1976) 67 10, ll. 12, 13. 14, 15. 16. 17. 18. 19, October 27, 1977 Dijkstra E.W., Notes on Structured Programming, in Dahl,0O., Dijkstra, E.W. and Hoare, C.A.R. in Structured Programming, Academic Press (1972) Dugaiczyk, A., Boyer, H. and Goodman, H., Ligation of EcoR1l Endonuclease- generated DNA Fragments into Linear and Circular Structures, J. Mol. Bio. 96:171-184 (1975) Ehrlich §.D., Bursztyn-Pettegrew H., Stroynowski I. and Lederberg J., Expression of the thymidylate synthetase gene of the Bacillus subtilis bacteriophage Phi-3-T in Escherichia coli, Proceedings of the National Academy of Sciences USA, 73:4145-4149 (1976) Evans 7.G., A Program for the Solution of Geometric Analogy Intelligence Test Questions, in First Order Mathematical Logic, Margaris, A., Blaisdell Publishing Company, Waltham, Mass. (1967) Feitelson J.S. and Stefik M.J., A Case Study of the Reasoning in a Genetics Experiment, Heuristic Programming Project Working Paper 77-18, Computer Science Department, Stanford University (1977) Fikes R.E. and Nilsson N.J., Learning and Executing Generalized Robot. Plans, Artificial Intelligence, 3:251~288 (1972) Fikes R.E. and Nilsson N.Jd., STRIPS: A New Approach to Applications of Theorem Proving to Problem Solving, Artificial Intelligence 2:189-208 (1971) Feigenbaum E.A., The Art of Artificial Intelligence: I, Themes and Case Studies of Knowledge Engineering, 5IJCAI:1014-1029 (1977) Gerhart S.L., What Goes Down Should Also Come Up: Some Issues about Abstraction, 5th Texas Conference on Computing Systems (1976) Gerhart S. L., Knowledge About Programs: A Model and Case Study, Proc. of Intl. Conf. on Reliable Software, Los Angeles (1975) 68 20. 21. 22, 23. 24, 25. 27. 28. 29. 30. 31. October 27, 1977 Goldberg A. and Kay A., SMALLTALK-72 Instruction Manual, Xerox Corporation (1976) Goldstein I.P. and Roberts R.B., NUDGE, A Knowledge-based Scheduling Program, 5IJCAI: 257-263 (1977) Hayes-Roth F, and Lesser V.R., Focus of Attention in the Hearsay- II Speech Understanding System, 5IJCAI:27-35 (1977) Kling R.E,, Reasoning By Analogy With Applications To Heuristic Problem Solving: A Case Study, Stanford Computer Science Department Report No. CS-216 (1971) Martin N., Friedland P., King J. amd Stefik M., Knowledge Base Management for Experiment Planning in Molecular Genetics, 5IJCAT: 882-887 (1977) McDermott D., Vocabularies for Problem Solver State Descriptions, SIJCAI:229-234 (1977) Miller M.L. and Goldstein I.P., Structured Planning and Debugging, SIJCAI:773-779 (1977) Minsky M.A., A Framework for Representing Knowledge, in Winston P (ed) The Psychology of Computer Vision, New York: McGraw-Hill -(1975) Notani, N.K. and Setlow, J.K., Mechanism of Bacterial Tranformation and Transduction, in Progress in Nucleic Acid Research and Molecular Biology 14:39-100 (1974) Platt J.R., Strong Inference, Science 146:347 (1964) Rieger C. and Grinberg M., The Declarative Representation and Procedural Simulation of Causality in Physical Mechanisms, DIJCAI: 250-256 (1977) Roberts, R.J, CRC Critical Reviews of Biochemistry 4 :123-164 (1976) 69 32. 33. 34, 35. 36. 37. 38. 39. 40, 4l. October 27, 1977 Sacerdoti E.D,.,. The Nonlinear Nature of Plans, 4IJCAI,:206-214 (1975) Sacerdoti E.D., Planning in a Hierachy of Abstraction Spaces, 3IJCAI: 412-422 (1973) Schaffner K., Logic of Discovery and Justification in Regulatory Genetics, Stud. Hist. Phil. Sci. 4:349-385 (1974) Scheller R.H., Dickerson R.E., Boyer H.W., Riggs A.D. and Itakura K., Chemical Synthesis of Restriction Enzyme Recognition Sites Useful for Cloning, Science 196:177-180 (1977) Stefik M.J. end Martin N., A Review of Knowledge Based Problem Solving as A Basis for a Genetics Experiment Designing System, Stanford Computer Science Department. Report No. STAN-CS-77-596 (1977) Sussman G.J., The Virtuous Nature of Bugs, Proceedings of the AISB Summer Conference (1974) Ullrich A., Shine J., Chirgwin J., Pictet R., Tischer E., Rutter W.J., and Goodman H.M., Rat Insulin Genes: Construction of Plasmids Containing the Coding Sequences, Science 196:1313-1319 (1977) Walker D.E., Paxton W.H., Grosz B.J., Hendrix G.G., Robinson A.E., Robinson J.J., Slocum J., Procedures for Integrating Knowledge in a Speech Understanding System, 5IJCAI:36-42 (1977) Wipke W.T., Computer-Assisted Three Dimensional Synthetic Analysis, in Computer Representation and Manipulation of Chemical Information, Wipke W.T. et.al. (eds.) John Wiley (1974) Woods B., What's in a Link?, in Bobrow and Collins (eds.) Representation and Understanding (1975) 70