human donor splice site sequence logo

A Glossary for Biological Information Theory and the Delila System

by Tom Schneider and Karen Lewis

gumball machine

  • absolute coordinate: A number (usually integer) that describes a specific position on a nucleic acid or protein sequence. An example of using two absolute coordinates in Delila instructions is:
    get from 1 to 6;
    The numerals 1 and 6 are absolute coordinates. See also: relative coordinate.
  • acceptor splice site: The binding site of the spliceosome on the 3' side of an intron and the 5' side of an exon. This term is preferred over "3' site" because there can be multiple acceptor sites, in which case "3' site" is ambiguous. Also, one would have to refer to the 3' site on the 5' side of an exon, which is confusing. Mechanistically, an acceptor site defines the beginning of the exon, not the other way around. See
  • acronymology: The study of words (as radar, snafu) formed from the initial letter or letters of each of the successive parts or major parts of a compound term.
  • administrivia: [Pronunciation: combine administ[ration] and trivia. Function: noun. Etymology: coined by TD Schneider. Date: before 2000] administritrative trivia
  • after state (after sphere, after): the low energy state of a molecular machine after it has made a choice while dissipating energy. This corresponds to the state of a receiver in a communications system after it has selected a symbol from the incoming message while dissipating the energy of the message symbol. The state can be represented as a sphere in a high dimensional space. See also: Shannon sphere, gumball machine, channel capacity.
  • alignment of all 17 bacteriphage T7 promoters above a sequence logo of them
    Alignment of all 17 bacteriphage T7 promoters
    above a sequence logo of them. The columns are: This figure was made using Delila programs,
    especially alist, makelogo and the alo script.
    The image is available in PostScript:
    and the Delila instructions are t7-aloall.inst for the T7 genome.
    alignment (align): a set of binding site or protein sequences can be brought into register so that a biological feature of interest is emphasized. A good criterion for finding an alignment is to maximize the information content of the set. This can be done for nucleic acid sequences by using the malign program. See also:
  • before state (before sphere, before): the high energy state of a molecular machine before it makes a choice. This corresponds to the state of a receiver in a communications system before it has selected a symbol from the incoming message. The state can be represented as a sphere in a high dimensional space. See also: Shannon sphere, gumball machine, channel capacity.
  • binding site: the place on a molecule that a recognizer (protein or macromolecular complex) binds. In this glossary, we will usually consider nucleic acid binding sites. A classic example is the set of binding sites for the bacteriophage Lambda Repressor (cI) protein on DNA (M. Ptashne, How eukaryotic transcriptional activators work, Nature, 335, 683-689, 1988). These happen to be the same as the binding sites for the Lambda cro protein. (The text mentioned in the figure is Sequence Logos: A Powerful Yet Simple, Tool.) See also
  • binding site symmetry: binding sites on nucleic acids have three kinds of asymmetry and symmetry: Placement of zero coordinate: For consistency, one can place the zero coordinate on a binding site according to its symmetry and some simple rules. Within the Delila system, the instshift program makes readjusting the zero coordinate easy.
    The Symmetry Paradox: Note that specific individual sites may not be symmetrical (i.e. completely self-complementary) even though the set of all sites are bound symmetrically. This raises an odd experimental problem. How do we know that a site is symmetric when bound by a dimeric protein if each individual site has variation on the two sides? If we assume that the site is symmetrical, then we would write Delila instructions for both the sequence and its complement. The resulting sequence logo will, by definition be symmetrical. If, on the other hand, we write the instructions so as to take only one orientation from each sequence, perhaps arbitrarily, then by definition the logo will be asymmetrical. That is, one gets the output of what one puts in. This is a serious philosophical and practical problem for creating good models of binding sites. One solution would be to use a model that has the maximum information content although this may be difficult to determine in many cases because of small sample sizes. Another solution is to orient the sites by some biological criterion, such as the direction of transcription controlled by an activator. See also:
  • bit: A binary digit, or the amount of information required to distinguish between two equally likely possibilities or choices. If I tell you that a coin is 'heads' then you learn one bit of information. It's like a knife slice between the possibilities:
    tail of coin head of coin
    Likewise, if a protein picks one of the 4 bases, then it makes a two bit choice.

    The 4 basepairs of DNA. Choosing one of 4 objects takes 2 bits.

    For 8 things it takes 3 bits. In simple cases the number of bits is the log base 2 of the number of choices or messages M: bits = log2M. Claude Shannon figured out how to compute the average information when the choices are not equally likely. The reason for using this measure is that when two communication systems are independent, the number of bits is additive. The log is the only mathematical measure that has this property! Both of the properties of averaging and additivity are important for sequence logos and sequence walkers. Even in the early days of computers and information theory people recognized that there were already two definitions of bit and that nothing could be done about it. The most common definition is 'binary digit', usually a 0 or a 1 in a computer. This definition allows only for two integer values. The definition that Shannon came up with is an average number of bits that describes an entire communication message (or, in molecular biology, a set of aligned protein sequences or nucleic-acid binding sites). This latter definition allows for real numbers. Fortunately the two definitions can be distinguished by context.
    See also:
  • BITCS: Biological Information Theory and Chowder Society. An eclectic group of folks from around the planet who are interested in the application of information theory in biology. Discussions are held on and there is a FAQ available.
  • blind alley: Avoiding blind alleys is hard. How do you know a path is blind without going into it? If you are a true explorer, you cannot trust another explorer's word that a certain way is blocked - maybe there is a way through that you will see. The most important thing is to go into interesting paths and explore them. It is important to be able to identify a path as a dead end (for you) and then to WALK OUT again. People usually just hang around and get stuck with lots of bad ideas in their heads. One example is to think that one's model is reality. See: pitfall and pitfalls in molecular information theory.
  • Boltzmann: Ludwig Boltzmann was a famous thermodynamicist who recognized that entropy is a measure of the number of ways (W) that energy can be configured in a system:
    S = K log W
    This formula is on Boltzmann's Tomb in the Zentralfriedhof (Central Cemetary). Vienna, Austria.Bust of Ludwig Boltzmann in Vienna, taken by Thomas
Schneider in 2002. See also:
  • book: A collection of DNA sequences in the Delila system. Books are usually created by the delila program, but can also be created by dbbk, rawbk, and makebk. The unique feature of Delila books is that they carry a coordinate system that defines the coordinates of each base in the book. This makes the Delila program powerful because one can use Delila to extract parts of sequences and maintain the original coordinates. See also library.
  • box: Poor Terminology! A region of sequence with a particular function. A sequence logo of a binding site will often reveal that there is significant sequence conservation `outside' the box. The term `core' is sometimes used to acknowledge this, but sequence logos reveal that the division is an arbitrary convention and therefore not biologically meaningful. Recommendation: replace this concept with binding site for nucleic acids or `motif' for proteins. Example: In the paper "Ordered and sequential binding of DnaA protein to oriC, the chromosomal origin of Escherichia coli". Margulies C, Kaguni JM. J Biol Chem 1996 Jul 19;271(29):17035-40, the authors use the conventional model that DnaA binds to 9 bases and they call the sites "boxes". However, in the paper they demonstrate that there are effects of the sequence outside the "box", which demonstrates that the "box" is an artifact.
  • byte: A binary string consisting of 8 bits.
  • certainty: 'Certainty' is not defined in information theory. However, Claude Shannon apparently discovered that one can measure uncertainty. By implication, there is no measure for 'certainty'. The best one can have is a decrease of uncertainty, and this is Shannon's information measure. The uncertainty before an event (e.g. receiving a symbol) less the equivocation (uncertainty after the event) is the information. Since there is always thermal noise, there is always equivocation, so there is never absolute certainty. See: uncertainty.
  • channel capacity, channel capacity theorem: The maximum information, in bits per second, that a communications channel can handle is:
    Shannon channel capacity equation: C equals W log base 2
of P over N plus 1 bits per second.
    where W is the bandwidth (cycles per second = hertz), P is the received power (joules per second) and N is the noise (joules per second). Shannon derived this formula by realizing that each received message can be represented as a sphere in a high dimensional space. The maximum number of messages is determined by the diameter of these spheres and the available space. The diameter of the spheres is determined by the noise and the available space is a sphere determined by the total power and the noise. Shannon realized that by dividing the volume of the larger sphere by the volume of the smaller message spheres, one would obtain the maximum number of messages. The logarithm (base 2) of this number is the channel capacity. In the formula, the signal-to-noise ratio is P/N. Shannon's channel capacity theorem states that if one attempts to transmit information at a rate R greater than C only at best C bits per second will be received. On the other hand if R is less than or equal to C then one may have as few errors as desired, so long as the channel is properly coded. As a consequence of this theorem, many methods of coding have been derived, and as a result we now have satellite communications, the internet, CDs, DVD's and wireless communications. A similar formula applies to biology.
    See also:
  • choice: The process whereby a living being (or part of one) discriminates between two or more symbols. For example, the EcoRI restriction enzyme binds to the pattern 5' GAATTC 3' in DNA. It avoids all other 6 long sequences. If you mix the enzyme with DNA, the DNA is cut between the G and A. That is, it is a molecule that picks GAATTC from all other patterns. It makes choices. Furthermore you can measure the number of choices in bits: 12 bits. See also:
  • code (coding, coding theory): Coding is the representation of a message into a form suitable for transmission over a communications line. This protects the message from noise. Since messages can be represented by points in a high dimensional space (the first bit is the first dimension, the second bit is the second dimension, etc., see message), the coding corresponds to the placement of the messages relative to each other in the high dimensional space. This concept is from Shannon's 1949 paper. When a message has been received, it has been distorted by thermal noise, and in the high dimensional space the noise distorts the initial transmitted message point in all directions evenly. The final result is that each received message is represented by a point somewhere on a sphere. Decoding the message corresponds to finding the nearest sphere center. Picking a code corresponds to figuring out how the spheres should be placed relative to each other so that they are distinguishable by not overlapping. This situation can be represented by a gumball machine. Shannon's famous work on information theory was frustrating in the sense that he proved that codes exist that can reduce error rates to as low as one may desire (since at high dimensions the spheres become sharp edged), but he did not say how this could be accomplished. Fortunately a large effort by many people established many kinds of communications codes, and of course the development of electronic chips allows decoding in a small device. The result is that we now have many means of clear communications, such as CDs, MP3, DVD, the internet, and digital wireless cell phones. One of the most famous coding theorists was Hamming. An example of a simple code that protects a message against error is the parity_bit. Codes exist in biology and molecular biology not only in the genetic code but also there must be a code for every specific interaction made by molecular machines. In many of these cases the spheres represent states of molecules instead of messages. See also:
  • communication: for information theory, communication is a process in which the state at a transmitter, a source of information, is reproduced with some errors at a receiver. The errors are caused by noise in the communications channel.
  • complexity: Poor Terminology! Like `specificity', the term `complexity' appears in many scientific papers, but it is not always well defined. (See however M. Li and P. Vitanyi, A Introduction to Kolmogorov Complexity and Its Applications, second edition, Springer-Verlag, New York, ISBN 0-387-94868-6, 1997) When one comes across a proposed use in the literature one can unveil this difficulty by asking: How would I measure this complexity? What are the units of complexity? Recommendation: use Shannon's information measure or explain why Shannon's measure does not cover what you are interested in measuring. Then give a precise, practical definition.
  • consensus sequence (consensus): Poor Terminology! The simplest form of a consensus sequence is created by picking the most frequent base at some position in a set of aligned DNA, RNA or protein sequences such as binding sites. The process of creating a consensus destroys the frequency information and leads to many errors in interpreting sequences. It is one of the worst pitfalls in molecular biology. Suppose a position in a binding site had 75% A. The consensus would be A. Later, after having forgotten the origin of the consensus while trying to make a prediction, one would be wrong 25% of the time. If this is done over all the positions of a binding site, most predicted sites can be wrong! For example, in Rogan and Schneider (1995) a case is shown where a patient was misdiagnosed because a consensus sequence was used to interpret a sequence change in a splice junction. Figure 2 of the sequence walker paper shows a Fis binding site that had been missed because it did not fit a consensus model. Recommendation: one can entirely replace this concept with sequence logos and sequence walkers. See also
  • coordinate system of sequences: A coordinate system is the numbering system of a nucleic acid or protein sequence. Coordinate systems in primary databases such as GenBank and PIR are usually 1-to-n, where n is the length of the sequence, so they are not recorded in the database. However, in the Delila system, one can extract sequence fragments from a larger database. If one does two extractions, then one can go slightly crazy trying to match up sequence coordinates if the numbering of the new sequence is still 1-to-n. The Delila system handles all continuous coordinate systems, both linear and circular, as described in LIBDEF, the definition of the DELILA database system. For example, on a circular sequence running from 1 to 100, the Delila instruction get from 10 to 90 direction -; will give a coordinate system that runs from 10 down to 1, and then continues from 100 down to 90.
  • core consensus: Poor Terminology! A core consensus is the strongly conserved portion of a binding site found by creating a consensus sequence. It is an arbitrary definition as can be seen from the examples in the sequence logo gallery. The sequence conservation, measured in bits of information, often follows the cosine waves that represent the twist of B-form DNA. This has been explained by noting that a protein bouncing in and out from DNA must evolve contacts. It is easier to evolve DNA contacts that are close to the protein than those that are further around the helix. Because the sequence conservation varies continuously, any cutoff or "core" is arbitrary. Recommendation: replace this concept with sequence logos and sequence walkers. See also:
  • Delila: stands for
    It is a language for extracting DNA fragments from a large collection of sequences, invented around 1980 (T. D. Schneider, G. D. Stormo, J. S. Haemer, and L. Gold", A design for computer nucleic-acid sequence storage, retrieval and manipulation, Nucl. Acids Res., 10: 3013-3024, 1982). The idea is that there is a large database containing all the sequences one would like, which we call a `library'. (It is amusing and appropriate that GenBank now resides at the National Library of Medicine in the National Center for Biotechnology Information!) One would like a particular subset of these sequences, so one writes up some instructions and gives them to the librarian, Delila, which returns a `book' containing just the sequences one wants for a particular analysis. So `Delila' also stands for the program that does the extraction (delila.p). Since it is easier to manipulate Delila instructions than to edit DNA sequences, one makes fewer mistakes in generating one's data set for analysis, and they are trivial to correct. Also, a number of programs create instructions, which provides a powerful means of sequence manipulation. One of Delila's strengths is that it can handle any continuous coordinate system. The `Delila system' refers to a set of programs that use these sequence subsets for molecular information theory analysis of binding sites and proteins. In the spring of 1999 Delila became capable of making sequence mutations, which can be displayed graphically along with sequence walkers on a lister map. A complete definition for the language is available (LIBDEF), although not all of it is implemented. There are also tutorials on building Delila libraries and using Delila instructions. A web-based Delila server is available.
  • Delila instructions: a set of detailed instructions for obtaining specific nucleic-acid sequences from a sequence database. The instructions are written in a computer language called Delila. There is a short tutorial on using Delila instructions.
  • digit: The set of symbols 0 to 9, specifying the choice of one thing in 10. Therefore, like the bit, a digit is a measure of an amount of information. While bits are determined by using log base 2, digits are determined by taking log base 10 (and adding 1). So the number 1000 is log101000 + 1 = 3 + 1 = 4 digits. It's not clear to my why one adds 1, but certainly the number of digits in a number follows this formula. However in the case of 1, we would like to say that there is one digit, so log101 + 1 = 0 + 1 = 1 digit.
  • donor splice site: The binding site of the spliceosome on the 5' side of an intron and the 3' side of an exon. This term is preferred over "5' site" because there can be multiple donor sites, in which case "5' site" is ambiguous. Also, one would have to refer to the 5' site on the 3' side of an exon, which is confusing. Mechanistically, a donor site defines the end of the exon, not the other way around. See Geometry for optimal bistate molecular machines.  Two
concentric circles are connected by a horizontal line
segment running from the outer circle on the left, tangent
to the inner circle in the middle and to the outer circle
on the right.  Behind the circles are concentric colors in
a spectrum running red at the center to purple on the edge
representing lower to higher energy.
  • efficiency: the amount of energy applied to a useful purpose in a system compared to the total energy dissipated. The Carnot efficiency functions between two temperatures. This is not appropriate for most biological systems since biological systems generally function at one temperature. An efficiency defined by Pierce and Cuttler in 1959 applies to isothermal systems. It is computed by dividing the information gained by the energy dissipated, when the energy has been converted to bits using the Second Law of Thermodynamics in the form Emin = kB T ln(2) = -q/R (joules per bit). See
  • entropy: A measure of the state of a system that can roughly be interpreted as the randomness of the energy in a system. Since the entropy concept in thermodynamics and chemistry has units of energy per temperature (Joules/Kelvin), while the uncertainty measure from Claude Shannon has units of bits per symbol, it is best to keep these concepts distinct. The Boltzmann form for entropy is:
    Boltzmann entropy equation: S equals the negative sum of
Kb from i equals 1 to Omega of P sub i natural log 2 p sub
i joules per Kelvin microstate.  Kb is Boltzmann's constant.
    while the Shannon form for uncertainty is:
    Shannon uncertainty equation: H equals the negative sum
from i equals 1 to M of P sub i log 2 p sub i bits per
    See also:
  • error: In communications, the substitution of one symbol for another in a received message caused by noise. Shannon's channel capacity theorem showed that it is possible to build systems with as low an error as desired, but one cannot avoid errors entirely.
  • Evolution of Biological Information: The information of patterns in nucleic acid binding sites can be measured as Rsequence (the area under a sequence logo). The amount of information needed to find the binding sites, Rfrequency, can be predicted from the size of the genome and number of binding sites. Rfrequency is fixed by the current physiology of an organism but Rsequence can vary. A computer simulation shows that the information in the binding sites (Rsequence) does indeed evolve toward the information needed to locate the binding sites (Rfrequency). See:
  • flip-flop: A flip-flop is a two-state device. A common example is a light switch. Flip-flops can store one bit of information. See also
  • frequency: A measured number of occurances of an event in a sample population. See also:
  • from: The 5' extent of the range of a binding site. For example in a Delila instruction one might have get from 50 -10 to same +5; the range runs from -10 to +5.
  • genetic control system: a set of one or more genes controlled by proteins or RNAs. There are thousands of examples. The most famous is the Lac repressor system, which was the first one understood. Jacob and Monod used elegant genetics to figure out how it worked. Basically a protein called the Lac Repressor binds to the DNA and so blocks transcription. Another famous system is the bacteriophage lambda cI repressor and cro system. There is a vast and rapidly growing literature as people figure out control systems in all the different organisms. Genetic control systems are involved in developmental biology, so the structure of animals and plants is determined by them. Many diseases are result of ruined or partially ruined controls. For example, 15% of all single point mutations that cause genetic diseases in humans are in splice junctions (donor and acceptor splice sites) ( Krawczak M, Reiss J, Cooper DN., Hum Genet. 1992 Sep-Oct;90(1-2):41-54.), which are part of a genetic control system that splices mRNA. Much of the rest of this web site has sequence logos for many genetic systems that we have analyzed. You can explore that too. See also:
  • genome: The complete genetic material of an organism. It can be either DNA or RNA. For example, the genome of the bacterium E. coli is about 4.7 million base pairs of DNA and has about 4,000 genes. By contrast a human has about 3 billion base pairs of DNA and has 20,000 to 25,000 genes. You can find the complete genomes of many organisms at GenBank. When computing the information needed to locate a set of binding sites in a genome, the number of positions that a protein or other molecule can bind is counted. This may not be the number of base pairs. See the discussion of Rfrequency for further explanation.
  • genomic skew: The frequencies of bases in the genome of an organism are not always equiprobable. For example, the composition can have high "GC" content relative to the "AT". If one makes a sequence logo, this can appear as a background information outside the binding sites. Many people immediately assume that it should be removed. This can generally be done by computing the genomic uncertainty and using that for Hbefore. However, this implies an interpretation of the phenomenon, and the cause of 'skew' is not understood. Some possibilities include strong biases in mutation or DNA repair. Alternatively, histone-like proteins could be binding all over the genome, in which case it would be inappropriate to remove the pattern, as it represents the actual information of a binding protein! For further discussion, see also:
  • gumball machine gumball machine: A model for the packing of Shannon spheres. Each gumball represents one possible message or one possible molecular state (an after sphere). The radius of the gumball represents the thermal noise. The balls are all enclosed inside a larger sphere (the before sphere) whose radius is determined from both the thermal noise and the power dissipated at the receiver (or by the molecule) while it selects that state. The way the spheres are packed relative to each other is the coding. See channel capacity and molecular machine capacity.
  • Richard W. Hamming: An engineer at Bell Labs in the 1940s who wrote the famous book "Coding and Information Theory" which explains coding theory. See also:
  • hypersphere: See Shannon sphere.
  • Independence: two variables are independent when a change of one of them does not alter the value of the other. We often assume that positions across a binding site are independent. This is a major assumption for sequence logos. The idea that different parts of a binding site are independent is a useful initial assumption. However, there is some literature on the subject of non-independence and Gary Stormo has written on it. In cases when there is enough data, one can test the assumption - in our 1992 paper on splicing we did and found no correlations for the acceptors and a small amount for donors. The likely reason for the general observation of independence in binding sites is pretty simple. The DNA or RNA lies in a groove on the surface of the molecule. Aside from neighboring bases, a complex system of correlations would be hard to evolve because it would require mechanisms running through the recognizer. So they evolve, for the most part, not to have correlations.

    Independence plays a vital role in information theory. When two communications channels are independent, the information of each can be added. Because he wanted an additive measure, Shannon demanded this and found (as others before him) that the log function has the necessary property. For example, suppose we have two channels, one has two symbols H and T (for heads and tails) and the other has four symbols a, c, g, t (for the four bases). Then the first can carry 1 bit (log22 = 1) and the second can carry 2 bits (log24 = 2). A combined channel has eight symbols (Ha, Hc, Hg, Ht, Ta, Tc, Tg, Tt) and can carry 3 bits (log28 = 3). That is, you can multiply the possibilities or you can add the logs. Independence plays an elegant role in Shannon's construction of the channel capacity in his 1949 paper. In this case it is worth noting that if two variables are independent, this can be represented geometrically as two orthogonal axes (at 90 degrees to each other). See also:
  • individual information: the information that a single binding site contributes to the sequence conservation of a set of binding sites. This can be graphically displayed by a sequence walker. It is computed as the decrease in surprisal between the before state and the after state. The technical name is Ri. See also: ridebate.
  • information: Information is measured as the decrease in uncertainty of a receiver or molecular machine in going from the before state to the after state.
    "In spite of this dependence on the coordinate system the entropy concept is as important in the continuous case as the discrete case. This is due to the fact that the derived concepts of information rate and channel capacity depend on the difference of two entropies and this difference does not depend on the coordinate frame, each of the two terms being changed by the same amount."
    --- Claude Shannon, A Mathematical Theory of Communication, Part III, section 20, number 3
    Information is usually measured in bits per second or bits per molecular machine operation. See also:
  • information theory: Information theory is a branch of mathematics founded by Claude Shannon in the 1940s. The theory addresses two aspects of communication: "How can we define and measure information?" and "What is the maximum information that can be sent through a communications channel?" (channel capacity). See also:
  • Geometry for optimal bistate molecular machines.  Two
concentric circles are connected by a horizontal line
segment running from the outer circle on the left, tangent
to the inner circle in the middle and to the outer circle
on the right.  Behind the circles are concentric colors in
a spectrum running red at the center to purple on the edge
representing lower to higher energy. isothermal efficiency: A measure of how a system uses energy when functioning at only one temperature.
    efficiency equation on blackboard
  • junk_DNA: regions of a genome for which we do not know a function. Calling large parts of the genome 'junk' is possibly the height of human egotism, unless it stands for J.U.N.K: Just Use Not Known.
  • leaky mutation: a weak mutation. For example, figure 2 in Rogan.Faux.Schneider1998.
  • library: A DNA sequence database in the Delila system. A library is created by running the catal program, which ensures that sequence fragments in the library do have the duplicated names. See also book.
  • lister feature: A graphical object marking a sequence on a lister map. Features are defined once and then may be used many times. Features can be either ASCII (i.e. text) strings or sequence walkers. In either case the lister program arranges the locations of the features so that they do not overlap. Programs that generate features are scan and search. See also: lister mark, lister map.
  • lister map: A graphical display of one or more sequences marked with protein translations, colored marks (generally arrows and boxes but also cyclic waves), ASCII features (such as footprinted regions, exons and RNA structures) and sequence walkers. The map is produced by the lister program. Some examples: See also:
  • lister mark or mark: A graphical object associated with a sequence on a lister map or sequence logo. Marks are defined entirely by coordinate position and do not displace features or other marks. They are placed at the time that the coordinate is encountered by lister and follow the postscript rule that younger marks are written on top of older ones. This allows one to place boxes around sequence walkers, for example, by placing the mark for a box before the walker coordinate. Marks must be given strictly in the order of the book sequences. When sequence polymorphisms or mutations are generated by the delila program, they are recorded using the marksdelila file. The live program generates cyclic marks along the sequence and can be used to indicate the face of the DNA or the reading frame. A user can also define their own kind of marks using PostScript. A set of marks can be merged with other marks using the mergemarks program. See also: lister feature, lister map, makelogo, marks.arrow, libdef examples of marksdelila.
  • logo: See: sequence logo.
  • map: See lister map.
  • Maxwell's demon: A mythical beast invented by James Maxwell in 1867. The demon supposedly violates the Second Law of Thermodynamics. However, a careful analysis from the perspective of molecular biology indicates that such a creature is not possible to construct given our current knowledge of atoms, myosin, actin and rhodopsin (see nano2).
  • meaning: In his 1948 paper, Shannon explicitly set aside value and meaning in his exposition of information theory:
    Frequently the messages have meaning; that is they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem.
    But what is meaning? Many have struggled with this question. I (TDS) thought of meaning as the interpretation of information by a being, but a clear exposition is given by Anthony Reading in Information 2012, 3, 635-643 When Information Conveys Meaning. His definition:
    Meaningful information is thus conceptualized here as patterns of matter and energy that have a tangible effect on the entities that detect them, either by changing their function, structure or behavior, while patterns of matter and energy that have no such effects are considered meaningless.
  • message: A message is a series of symbols chosen from a predefined alphabet. In molecular biology the term `message' usually refers to a messenger RNA. In molecular information theory, a message corresponds to an after state of a molecular machine. In information theory, Shannon proposed to represent a message as a point in a high dimensional space (see Shannon1949). For example, if we send three independent voltage pulses, their heights correspond to a point in three-dimensional space. A message consisting of 100 pulses corresponds to a point in 100 dimensional space. Starting from this concept, Shannon derived the channel capacity. See also:
  • mismatches: Poor Terminology! The number of mismatches is a count of the number of differences between a given sequence and a consensus sequence. For example, a friend wrote "This binding site has three mismatches in non-critical positions." If one wants to note that a position in a binding site has negative information in a sequence walker, then one can say that it has negative information! A base in a site could have a mismatch to the consensus and yet that base could contribute positive information. For example for a position that has 60% A, 30% T, 5% G, and 5% C the consensus base is A by two-fold, and yet a T in an individual binding site would contribute 2 + log20.30 = 0.26 bits. Chari (Krishnamachari Annangarachari) pointed out that the lesson is to "see things in totality, not in isolation". That is, only by noting the total distribution can we learn that the T contributes positively to the total information. He also pointed out that this is a lesson in "unity in diversity". The logo shows both the unity of the binding site and simultaneously shows its diversity. See also:
  • molecular biology: The study of biology at the molecular level. Molecular biologists have no fear of stealing from adjacent scientific fields. When they discovered the structure of DNA Watson and Crick used ideas from physics, genetics and biochemistry (already a conglomeration of biology and chemistry). This web site is all about stealing information theory and taming it for molecular biologists. See also: equation for MIT
  • molecular information theory: Information theory applied to molecular patterns and states. The more general term is Biological Information Theory = BIT, coined by John Garavelli. For a review see the nano2 paper. Google search for "molecular information theory".
  • molecular machine: The definition given in Channel Capacity of Molecular Machines is:
    1. A molecular machine is a single macromolecule or macromolecular complex.
    2. A molecular machine performs a specific function for a living system.
    3. A molecular machine is usually primed by an energy source.
    4. A molecular machine dissipates energy as it does something specific.
    5. A molecular machine `gains' information by selecting between two or more after states.
    6. Molecular machines are isothermal engines.
  • molecular machine capacity: The maximum information, in bits per molecular operation that a molecular machine can handle. When translated into molecular biology, Shannon's channel capacity theorem states that
    By increasing the number of independently moving parts that can interact cooperatively to make decisions, a molecular machine can reduce the error frequency (rate of incorrect choices) to whatever arbitrarily low level is required for survival of the organism, even when the machine operates near its capacity and dissipates small amounts of power.
    (quoted from page 112 of T. D. Schneider, J. Theor. Biol., 83-123:, 112, 1991.) This theorem explains the precision found in molecular biology, such as the ability of the restriction enzyme EcoRI to recognize 5' GAATTC 3' while ignoring all other sites. See the related channel capacity. The derivation is in T. D. Schneider, Theory of Molecular Machines. I. Channel Capacity of Molecular Machines. J. Theor. Biol., 148:, 83-123, 1991.
  • molecular machine operation: The thermodynamic process in which a molecular machine changes from the high energy before state to a low energy after state. There are four standard examples:
  • motif: See pattern.
  • mutation: a nucleic-acid sequence change that affects biological function, for example by changing the information content of a binding site. A simple example is a primary splice site mutation. Delila instructions can be used to create mutations, and sequence walkers can be used to distinguish mutations from polymorphisms. Interestingly, a `mutation' depends on the function that one is considering. For example, one could have two overlapping binding sites. A sequence change can blow one away and leave the other one untouched (ie the Ri doesn't change). An interesting case of a cryptic splice acceptor next to a normal acceptor demonstrates how a single base change can have opposite effects on two splice sites. Another lovely example is the ABCR mutation. See also: polymorphism, leaky_mutation and a discussion on mutations and polymorphisms.
  • nanotechnology: Technology on the nanometer scale. The original definition is technology that is built from single atoms and which depends on individual atoms for function. An example is an enzyme. If you mutate the enzyme's gene, the modified enzyme may or may not function. In contrast, if you remove a few atoms from a hammer, it still will work just as well. This is an important distinction that has generally been lost as the hype about nanotechnology and it is used as a buzz word for 'small' instead of a distinctly different technology. Fortunately real nanotechnologies are in the works. See:
  • nat: Natural units for information or uncertainty are given in nats or nits. See nit for more.
  • negentropy: Poor Terminology! In his book "What is life?" Erwin Schrödinger said that "What an organism feeds upon is negative entropy." The term negentropy was defined by Brillouin (L. Brillouin, Science and Information Theory, second, Academic Press, Inc., New York, 1962, page 116) as `negative entropy', N = -S. Supposedly living creatures feed on `negentropy' from the sun. However it is impossible for entropy to be negative, so `negentropy' is always a negative quantity. The easiest way to see this is to consider the statistical-mechanics (Boltzmann) form of the entropy equation:
    Boltzmann entropy equation: S equals the negative sum of
Kb from i equals 1 to Omega of P sub i natural log 2 p sub
i joules per Kelvin microstate.  Kb is Boltzmann's constant.
    where kb is Boltzmann's constant, The Greek symbol omega. is the number of microstates of the system and Pi is the probability of microstate i. Unless one wishes to consider imaginary probabilities (!) it can be proven that S is positive or zero. Rather than saying `negentropy' or `negative entropy', it is more clear to note that when a system dissipates energy to its surroundings, its entropy decreases. So it is better to refer to -delta S (a negative change in entropy). Recommendation: replace this concept with `decrease in entropy'.

    The term `feeding on negentropy' is misleading because organisms eat physical matter (of course) that supplies them with energy. The energy is used to put molecules into the before state (H(X)) from which they THEN can make selections thereby gaining information by the molecule dropping to one of several possible lower energy after states (H(X|Y)). The potential energy in a sugar molecule and then ATP isn't directed initially to any particular choice and so isn't associated with any information process until it is used for one. The energy drop and the ultimate usefulness of energy comes only from it being spread out - to increase the entropy of the surroundings.

  • nit: Natural units for information or uncertainty are given in nits. If there are M messages, then ln(M) nits are required to select one of them, where ln is the natural logarithm with base e (=2.71828...). Natural units are used in thermodynamics where they simplify the mathematics. However nits are awkward to use because results are almost never integers. In contrast, the bit unit is easy to use because many results are integer (e.g. log2 32 = 5) and these are easy to memorize. Using the relationship ln x / ln 2 = log2 x allows one to present all results in bits.
    See also:
  • noise: A physical process that interferes with transmission of a message. Shannon pointed out that the worst kind of noise has a Gaussian distribution Shannon1949. Since thermal noise is always present in practical systems, received messages will always have some probability of having errors.
  • parameter file: Many Delila programs have parameter files. These are always simple text files, which is a robust method that will work on any computer system. Details on how to create the parameter file are always given on the manual page for the program. It is usually easiest to start from an example (also given on the manual page) and modify it. Parameters are given on individual text lines. If the entire line is not a parameter, then any text after the parameter is ignored and serves as a comment. For example, the program alist is controled by a file `alistp', which stands for alist-parameters. Some parameter files now indicate the version number of the program that they work with, and some programs are now able to use this to upgrade the parameter file automatically. See also the shell program.
  • parity bit: A parity bit determines a code in which one data bit is set to either 0 or 1 so as to always make a transmitted binary word contain an even or odd number of 1s. The receiver can then count the number of 1's to determine if there was a single error. This code can only be used to detect an odd number of errors but cannot be used to correct any error. Unfortunately for molecular biologists, the now-universal method for coding characters, 7 bit ASCII words, assigns to the symbols for the nucleotide bases A, C and G only a one bit difference between A and C and a one bit difference between C and G:
    A: 1018 = 10000012
    C: 1038 = 10000112
    G: 1078 = 10001112
    T: 1248 = 10101002
    For example, this choice could cause errors during transmission of DNA sequences to the international sequence repository, GenBank. If we add a parity bit on the front to make an even parity code (one byte long), the situation is improved and more resistant to noise because a single error will be detected when the number of 1s is odd:
    A: 1018 = 010000012
    C: 3038 = 110000112
    G: 1078 = 010001112
    T: 3248 = 110101002
  • Pascal's Triangle: A triangle of numbers
       1 1
      1 2 1
     1 3 3 1
    1 4 6 4 1
    in which each successive row is determined by adding the two numbers to the upper left and right of a number. The resulting distribution is a binomial distribution and in the limit of infinite rows, it approaches a Gaussian distribution. As pointed out by
    Edward Tarte, each number represents the number of ways that one can reach that point in the triangle. The concept of `ways' is, of course, the basis of the Second Law of Thermodynamics. It also shows that the Gaussian distribution is the `worst' kind of noise, as Shannon pointed out in 1949, since all paths are taken without discrimination. See also:
  • pattern: see sequence_pattern
  • John R. Pierce: An engineer at Bell Labs in the 1940s who wrote an excellent introductory book about information theory: An Introduction to Information Theory: Symbols, Signals and Noise. Though one would think it is out of date, it is still more clear and yet complete than anything else I have seen. He gave a wonderful talk at Bell Labs in December, 1951 entitled CREATIVE THINKING.
  • pitfall: An intellectual error that traps a researcher, perhaps forever. See the pitfalls web page for examples. See also blind alley and La Brea tar pits.
  • plörk, plurk: [Pronunciation: plûrk as in `work' and `urge'. Function: noun. Etymology: English from play and work, coined by TD Schneider; umlaut suggested by HA Schneider to ensure correct pronunciation. The umlaut is also a reference to their Austrian heritage. Alternative spelling suggested by LR Schneider Engle: plurk. Date: 2000. Earlier independent origin in 1997 by Teri-E Belf in the book Simply Live It Up: Brief Solutions] Play-work. Plörk is what scientists do. It is the enthusiastic, energetic application of oneself to the task at hand as a child excitedly plays; it is the intense arduous, meticulous work of an artist on their life-long masterpiece; it is joyful work.
    2004 May 13: The 1997 book Simply Live It Up: Brief Solutions by Teri-E Belf introduces the term plurk in three chapters, starting on page 143.
  • polymorphism: a DNA sequence change that does not affect biological function, or affects it non-lethally. Delila instructions can be used to create polymorphisms, and sequence walkers can be used to distinguish polymorphisms from mutations. See also: mutation and a discussion on mutations and polymorphisms.
  • position: a number defining where one is relative to the zero coordinate of a binding site.
  • probability: The number of occurances of an event in the entire population. See also: frequency.
  • qubit: A "quantum bit" is a device that can store not only two states, as a classical bit, but also, as in quantum mechanics, a superposition of two states. An example would be an electron in a magnetic field being either 'up', 'down' or a superposition of these states. Supposedly one could have an electron 'entangled' with another electron and do computation using them. I am not an expert in this field, but it is appropriate to at least mention it. This glossary is for Molecular Information Theory, not quantum information theory. The strong distinction between these two topics is that for quantum computers people want to avoid 'decoherence' because this destroys quantum computations. That is, they wish to avoid thermal noise. In the long run, this is basically impossible, but they might be able to do it for a long enough time to sneak a useful computation in. (It is impossible because of the third law of thermodynamics which says one cannot extract all the heat from a system to get it to absolute zero. However one might extract enough that there are only a few phonons of sound bouncing around.) In contrast, molecules in living things are totally bombarded by thermal noise (a "thermal maelstrom", ccmm) in which decoherence would happen quickly. So in the field of molecular information theory and biology in general on this planet, which is at 300K, it seems unlikely that one will find qubits in biological systems. See also: diagram of quincunx
  • quincunx: a device invented by Galton that demonstrates how the Gaussian distribution is generated. The device, shown to the right (source: Wikipedia), has balls starting from a single point that traverse through a field of pins and are collected into a series of slots at the bottom. The path of each ball has two reasonably random possibilities at each pin, so the final position of the balls forms a binomial distribution, which is a good approximation to a Gaussian distribution when there are many slots for collecting the balls. When multiple Gaussian distributions are joined at right angles, they form a sphere. The Amazing Normal Distribution Function - image from
the YouTube video showing a quincunx building a binormal
distribution in seconds. See also:
  • range: The region of positions ("from" to "to") that covers the site. The range can be chosen as the region which has significant sequence conservation above the fluctuation of conservation caused by small samples of sequence. Generally this can be done by looking at a sequence logo. See R. M. Stephens and T. D. Schneider, Features of spliceosome evolution and function inferred from an analysis of the information at human splice sites, J. Mol. Biol., 228: 1124-1136, 1992.
  • recognizer: A general term for a macromolecule that recognizes a specific pattern on a nucleic acid. This includes proteins such as transcription factors and protein/RNA complexes such as ribosomes and spliceosomes. See: binding site.
  • relative coordinate: A number (usually integer) that describes a specific position on a nucleic acid or protein sequence, as an offset from an absolute coordinate. An example of using relative coordinates in Delila instructions is: "get from 3 -2 to 3 +2;" The numerals -2 and +2 are coordinates relative to the absolute coordinate +3.
  • Rfrequency: The amount of information needed to find a set of binding sites out of all the possible sites in the genome. If the genome has G possible binding sites and γ binding sites, then Rfrequency = Hbefore - Hafter = log2G - log2γ = log2G/γ = -log2γ/G bits per site. Note that γ/G is the frequency of the sites in the genome. Rfrequency predicts the expected information in a binding site, Rsequence.

    Why is G not necessarily the genome size? G is the number of distinct places that a protein (for example) can bind to the genome. So the asymmetric bacteriophage T7 RNA polymerase can bind in two ways to each base pair. For example, the E. coli genome is a 4.7x106bp circle. That means that the polymerase could bind in G = 2x4.7x106 ways. In thermodynamic terms, G is the number of microstates possible (in the before state).

    For a ribosome, the genome is pretty much only transcribed once and not the complement (which would cause trouble because the complements would bind together blocking translation among other effects). So there are only G = 4.7x106 ways in that case.

    If a protein has dimeric symmetry (e.g. LacI, TrpR, LexA, etc) then it has two ways to bind, and G = 2x4.7x106. But there are also twice as many ways to bind at each binding sites (two at each base pair) so these extra factors of 2 cancel in the computation of Rf.

    See also Schneider1986.
  • Ri: Short hand notation for individual information. Following Claude Shannon, the `R' stands for a `rate of information, transmission'. For molecular biologists this is usually bits per base or bits per amino acid. The `i' stands for `individual'.
  • Ribl or Ri(b,l): a weight matrix constructed using individual information theory that is a model for a binding site.
  • Rsequence: The total amount of information conserved in a binding site, represented as the area under the sequence logo. Rsequence is computed as the recognizer's uncertainty before binding minus its uncertainty after binding. The uncertainty before binding could be 2 bits per base if you adhere to the philosophy that the recognizer cannot account for base composition when it is not bound. Alternatively, The uncertainty before binding could be the uncertainty computed from the genome (for DNA recognizers) or the parts of the genome that are available to the recognizer (for RNA recognizers). Using the genomic uncertainty will remove apparent background information, but this may not be the right thing to do since that might actually be information from binding proteins. The uncertainty after binding is computed from an aligned set of binding sites. It is not appropriate to compute the information in binding sites using the likelyhood function, as Gary Stormo does, because then you won't get bits since you can get results that claim to have more than 2 bits of information for positions that are entirely one base. Of course that violates the definition of information from Shannon. It is important to correct for small numbers of sequences. See also information, uncertainty, Rfrequency, recognizer, Schneider1986.
  • science:
    If you don't make mistakes, you're doing it wrong.
    If you don't correct those mistakes, you're doing it really wrong.
    If you can't accept that you are mistaken, you're not doing it at all.
  • Score: Poor Terminology! Many methods in "bioinformatics" give results as "scores". These ratings can be multiplied by an arbitrary constant and one still can compare them. In contrast, information is measured in bits, and this cannot be multiplied by an arbitrary constant and still retain the same units of measure. Scores cannot be compared between different binding sites, whereas it is reasonable to compare bits. For example, it is interesting that donor splice sites do not have the same information as acceptor splice sites (Stephens and Schneider J. Mol. Biol., 228: 1124-1136, 1992).

     Bits are not 'scores'. Would you call measured binding energies scores? No? Then notice that energy is closely related by an inequality to information, see: T. D. Schneider Theory of Molecular Machines. II. Energy Dissipation from Molecular Machines J. Theor. Biol., 148 (1): 125-137, 1991. That's a version of the Second Law of Thermodynamics. Bits are on an absolute scale that is directly related to the physical universe. Also, scores generally are integers but bits are real numbers.

    The distinction between scores and bits becomes critical when one is interested in computing the efficiency of a molecular system. Scores cannot be used, but bits can be compared to the energy by using the second law as an ideal conversion factor. See the discovery 70% efficiency of bistate molecular machines explained by information theory, high dimensional geometry and evolutionary convergence.

    See also sequence_conservation.

  • Second Law of Thermodynamics: The principle that the disorder of an isolated system (entropy) increases to a maximum. The Second Law appears in many surprisingly distinct forms.

    Transformations between these forms were described by Ed T. Jaynes ( in a beautiful paper (Jaynes, E. T., 1988, ``The Evolution of Carnot's Principle'' in Maximum-Entropy and Bayesian Methods in Science and Engineering, 1, G. J. Erickson and C. R. Smith (eds.), Kluwer, Dordrecht, p. 267; number 65,, postscript (56Kb) file.)

    The second law has many forms, and they do allow increase or decrease of entropy. In particular, dS >= dQ/T, the entropy change (increase) is always equal or greater than the heat input at some temperature T. Multiply both sides by -1 to get -dS <= -dQ/T, the entropy decrease is less than the heat dissipated (-dQ) at T. A snowflake is an example: heat leaves the region and the snowflake crystalizes into patterns.

    The relevant form for molecular information theory is Emin = kB T ln(2) = -q/R (joules per bit), where: kB is Boltzmann's constant, T is the absolute temperature, ln(2) is a constant that converts to bits, -q is the heat dissipated away from the molecular machine, and R is the information gained by the molecular machine. This form is derived in T. D. Schneider, Theory of Molecular Machines. II. Energy Dissipation from Molecular Machines J. Theor. Biol., 148: 125-137, 1991.

    Supposedly Maxwell's demon violates the Second Law, but if we approach the question from the viewpoint of modern molecular biology, the puzzles go away, as described in: T. D. Schneider, Sequence Logos, Machine/Channel Capacity, Maxwell's Demon, and Molecular Computers: a Review of the Theory of Molecular Machines Nanotechnology, 5, 1-18, 1994. is a light-hearted introduction to the "Mother of all Murphy's Laws":
    The Second Law of Thermodynamics!  (written in script)
    It clarifies an issue that is often incorrectly presented in discussions about entropy. See the other references by Frank Lambert below.

    See also:

  • Sequence conservation (conservation): Surprisingly, the degree of biological sequence conservation is neatly given in bits of information. One can envision that eventually all forms of biological conservation could be measured this way.

  • human donor splice site sequence logo sequence logo: A graphic representation of an aligned set of sequences, including DNA and RNA binding sites or protein sequences invented by Tom Schneider (owner of this website) and Mike Stephens. A logo displays the frequencies of bases or amino acids at each position, as the relative heights of letters, along with the degree of sequence conservation as the total height of a stack of letters, measured in bits of information. Subtle frequencies are not lost in the final product as they would be in a consensus sequence.
    Sequence logo for 38 LexA binding sites on E.  coli DNA
covering the range from the 5' side at -10 to the 3' side
at 11.  The vertical scale is from 0 to 2 bits.  The logo
letters approximately follow a sine wave running from 1 to
2 bits with wavelength 10.6 bases because LexA binds into
two adjacent major grooves of the DNA.
    ps, pdf
    The vertical scale is in bits, with a maximum of 2 bits possible at each position for DNA or RNA (with 4 bases, log24 = 2 bits per base) and log220 = 4.3 bits per amino acid for proteins. Note that sequence logos are an average picture of a set of binding sites (which is why logos can have several letters in each stack) while sequence walkers are the individuals that make up that average (which is why walkers have only one letter per position).
  • sequence pattern: A sequence pattern is defined by the nucleotide sequences of a set of aligned binding sites or by a common protein structure. In contrast, consensus sequences, sequence logos and sequence walkers are only models of the patterns found experimentally or in nature. Models do not capture everything in nature. For example, there might be correlations between two different positions in a binding site. A more sophisticated model might capture these but still not capture three-way correlations. It is impossible to make the more detailed model if there is not enough data. (In a more zen-like mood, we should note that everything we sense and observe is a model ...) See also:
  • sequence walker for human donor splice junctions sequence walker: A graphic representation of a single possible binding site, with the height of letters indicating how bases match the individual information weight matrix at each position. Bases that have positive values in the weight matrix are shown right-side up; bases that have negative values are shown upside down and below the "horizon". As in a sequence logo, the vertical scale is in bits; the maximum is 2 bits and the minimum is negative infinity. Bases that do not appear in the set of aligned sequences are shown negatively and in a black box. Bases that have negative values lower than can fit in the space available have a purple box. The zero coordinate is inside a rectangle which (in this case) runs from -3 to +2 bits in height. If the background of the rectangle is light green, the sequence has been evaluated as a binding site, while if it is pink it is not a binding site. Cover of Nucleic Acids Research November 1997 showing
sequence logos for human donor and acceptor splice junction
sites above 5 example sequence walkers for each splice
junction.See the sequence walker paper page.
  • Claude E. Shannon (April 30, 1916 - February 24, 2001): An engineer at Bell Labs in the 1940s who developed information theory. His most famous work is A Mathematical Theory of Communication, published in 1948. As a consequence of his work, we now have clear communication systems, including long distance voice phone calls, CDs without static, and the internet. Shannon was known to juggle and ride his unicycle in the halls of Bell Labs. See also:
  • Shannon Entropy: Poor Terminology! The story goes that Shannon didn't know what to call his measure and so asked the famous mathematician von Neumman. Von Neumann said he should call it the entropy because nobody knows what that is and so Shannon would have the advantage in every debate! This has led to much confusion in the literature because entropy has different units than uncertainty. It is the latter which is usually meant. If one does not use correct units, one will not get correct results. Recommendation: if you are making computations from symbols, always use the term uncertainty, with recommended units of bits per symbol. If you mean the entropy of a physical system, then use the term entropy, which has units of joules per kelvin (energy per temperature). See: The story is paraphrased from M. Tribus and E. C. McIrvine, "Energy and Information", Sci. Am., 225, (Note: the table of contents in this volume incorrectly lists this as volume 224), 3, 179-188, September, 1971. (PDF link) Here's the story:
    What's in a name? In the case of Shannon's measure the naming was not accidental. In 1961 one of us (Tribus) asked Shannon what he had thought about when he had finally confirmed his famous measure. Shannon replied: "My greatest concern was what to call it. I thought of calling it 'information,' but the word was overly used, so I decided to call it 'uncertainty.' When I discussed it with John von Neumann, he had a better idea. Von Neumann told me, 'You should call it entropy, for two reasons. In the first place your uncertainty function has been used in statistical mechanics under that name, so it already has a name. In the second place, and more important, no one knows what entropy really is, so in a debate you will always have the advantage.'
  • Shannon sphere: A sphere in a high dimensional space which represents either a single message of a communications system (after sphere) or the volume that contains all possible messages (before sphere) could be called a Shannon sphere, in honor of Claude Shannon who recognized its importance in information_theory. The radius of the smaller after spheres is determined by the ambient thermal noise, while that of the larger before sphere is determined by both the thermal noise and the signal power (signal-to-noise ratio), measured at the receiver. The logarithm of the number of small spheres that can fit into the larger sphere determines the channel capacity (See: Shannon1949). The high-dimensional packing of the spheres is the coding of the system.

    There are two ways to understand how the spheres come to be. Consider a digital message consisting of independent voltage pulses. The independent voltage values specify a point in a high dimensional space since independence is represented by coordinate axes set at right angles to each other. Thus three voltage pulses correspond to a point in a 3 dimensional space and 100 pulses correspond to a point in a 100 dimensional space. The first `non-Cartesian' way to understand the spheres is to note that thermal noise interferes with the initial message during transmission of the information such that the received point is dislocated from the initial point. Since noisy distortion can be in any direction, the set of all possible dislocations is a sphere. The second `Cartesian' method is to note that the sum of many small dislocations to each pulse, caused by thermal noise, gives a Gaussian distribution at the receiver. The probability that a received pulse is disturbed a distance x from the initial voltage is of the form p(x) ≈ e-x2. Disturbance of a second pulse will have the same form, p(y) ≈ e-y2. Since these are independent, the probability of both distortions is multiplied: p(x,y) = p(x) p(y). Combining equations, p(x,y) ≈ e-(x2 + y2) = e-r2, where r is the radial distance. If p(x,y) is a constant, the locus of all points enscribed by r is a circle. With more pulses the same argument holds, giving spheres in high dimensional space. Shannon used this construction in his channel capacity theorem. For a molecular machine containing n atoms there can be as many as 3n-6 independent components (degrees of freedom) so there can be 3n-6 dimensions. The velocity of these components corresponds to the voltage in a communication system and they are disturbed by thermal noise. Thus the state of a molecular machine can also be described by a sphere in a high dimensional velocity space. See also:
  • signal-to-noise ratio: The ratio between the received signal power and the noise at the receiver of a communications system. In molecular biology, the equivalent is the energy dissipated divided by the thermal noise.
  • site: See binding site.
  • skew, genomic: see genomic skew.
  • small sample correction: a correction to the Shannon uncertainty measure to account for the effects of small sample sizes. See:
  • specificity: Poor Terminology! The term is often ill defined. It has been used to refer to livers (tissue specificity), energy, binding constants, error rates, binding patterns (like what a sequence logo shows) and other mutually inconsistent concepts. Recommendation: use the appropriate precise term (energy, bits, information etc.) instead.
    The point I will try to make today is that one of the bigger problems I think a field like immunology has is really that we tend to use words that instill the feeling of understanding things but in fact only obfuscate things. And I just just pick one of these terms, um, specificity, you know we measure antibody activities in an Eliza so we stick something on plastic and hope the antibody sticks to the stuff that sticks on plastic and then we say we have a signal and this is important. Now when you measure most of these antibody qualities the binding quality is in the order of ten to the minus five to ten to the minus six molar. But when you then do measurements and assay antiviral or antibacterial protective antibodies you find that the avidities are in the order of ten to the minus nine molar. So a thousand to ten thousand times away from the scale you usually use.
    -- Rolf Zinkernagel, M.D., Ph.D., Zurich University and 1996 Nobel Laureate in a talk at NIH, "Anti-Viral Immunity and Vaccines" Wednesday, April 14, 2004. (3 minutes 10 seconds into the talk.) (search for Zinkernagel, direct video link)
  • surprisal: How surprised one would be by a single symbol in a stream of symbols. It is computed from the probability of the ith symbol, Pi, as ui = - log2Pi. For example, late at night, as I write this, the phone rarely rings so the probability of silence is close to 1 and the surprisal for silence is near zero. (If the probability of silence is 99% then usilence = - log20.99 = 0.01 bits per second, where the phone can ring only once per second.) On the other hand, a ring is rare so the surprisal for ringing is very high. (For example, if the probability of ringing is 1% per second then uring = - log20.01 = 6.64 bits per second.) The average of all surprisals over the entire signal is the uncertainty. (0.99 * 0.01 + 0.01 * 6.64 = 0.08 bits per second in this "phone"y example. ;-) The term comes from Myron Tribus' book "Thermostatics and Thermodynamics" (D. van Nostrand Company, Inc., Princeton, N. J., 1961).
  • symbol: two or more discrete physical states of a physical system associated with living beings. The states become separated from each other during the co-evolution of the beings and the symbols. Shannon's channel capacity theorem gurarantees that the states can be distinguished sufficiently for the survival of the organisms. See
  • symmetry: See binding site symmetry.
  • thermal noise: Thermal noise is caused by the random motion of molecules at any temperature above absolute zero Kelvin. Since the third law of thermodynamics prevents one from extracting all heat from a physical system, one cannot reach absolute zero and so cannot entirely avoid thermal noise. In 1928 Nyquist worked out the thermodynamics of noise in electrical systems and in a back-to-back paper Johnson demonstrated that the theory was correct. See:
  • to: The 3' extent of the range of a binding site. For example in a Delila instruction one might have get from 50 -10 to same +5; the range runs from -10 to +5.
  • tominology: [Pronunciation: tom-in-ology, Function: noun, Etymology: coined by TD Schneider, from the accidental combination of `Tom' and `terminology', Date: 2002 August 26]. Tom's obscure terminology, such as the word Rsequence, plurk, etc. Generally one should use the available terminology, but one should not fear inventing invent new terminology to describe new things.
  • uncertainty: A logarithmic measure of the average number of choices that a receiver or a molecular machine has available. The uncertainty is computed as:
    Shannon uncertainty equation: H equals the negative sum
from i equals 1 to M of P sub i log 2 p sub i bits per
    where Pi is the probability of the ith symbol and M is the number of symbols. Uncertainty is the average surprisal. The information is the difference between the uncertainty before and the uncertainty after symbol transmission. See also:
  • walker: See: sequence walker.
  • weight matrix: A two dimensional array of numbers that assigns to all possible sequences a weight. There are many methods for creating weight matrices:
  • zero coordinate (zero base, zero position): The position by which a set of binding sites is aligned. Not having a zero as part of your coordinate system is a bad idea thumb pointing down with a red sleeve because it makes computations tricky. On the positive side, having a zero coordinate allows one to precisely define the location of a binding site or other feature. In particular, it allows one to place a sequence walker. See also:
    Diamond shaped yellow street sign showing a man with a
shovel and a pile of dirt meaning that this page is under
construction Possible new terms to be added: Do you have a suggestion?

    Cover of the book Dictionary of Bioinformatics and
Computational Biology. Dictionary of Bioinformatics and Computational Biology, Hancock, John M. / Zvelebil, Marketa J. (eds.). (John Wiley & Sons, Inc., Hoboken, New Jersey, ISBN 0-471-43622-4, 2004.) I (Thomas D. Schneider) am one of the contributors. Many of the terms in this Glossary are in the Dictionary, and all of the terms in the Dictionary are in this Glossary.
    new as of 2004 August 5.

    Acknowledgments I thank Brent Jewett, Ryan Shultzaberger, Ilya Lyakhov, Jim Ellis, Krishnamachari Annangarachari (Chari), Danielle Needle, Guido De Mey (qubit), Bogdan S. Pecican (symbol, choice, certainty, communication), Mileidy Gonzalez (Rfrequency computation), Colin Kline (Ballarat, Victoria, Australia; probability and frequency) and David Dowe (Monash University; Clayton; Victoria 3168; Australia; nat) for useful questions, comments, corrections and suggestions.

    color bar

    Small icon for Theory of Molecular Machines: physics,
chemistry, biology, molecular biology, evolutionary theory,
genetic engineering, sequence logos, information theory,
electrical engineering, thermodynamics, statistical
mechanics, hypersphere packing, gumball machines, Maxwell's
Daemon, limits of computers
    Schneider Lab
    origin: 1999 April 15
    updated: version = 3.63 of glossary.html 2021 Dec 12

    color bar

    This is the bottom of the glossary. The blank lines above this point ensure that the last entry or entries start at the top of the page. If you have come here by using a pointer, then something is wrong, since there is no such definition. Please report the error to me. I'll need to know where you are so please send me the current url. If you are in frame mode, please use Show Only This Frame in your browser to get the correct url. I'll also need to know where you were so please back up with your browser to determine that and send me that url too. Thanks. Tom Schneider