Sequence motif

In biology, asequence motifis anucleotideoramino-acidsequencepatternthat is widespread and usually assumed to be related tobiological functionof the macromolecule. For example, anN-glycosylationsite motif can be defined asAsn, followed by anything but Pro, followed by either Ser or Thr, followed by anything but Pro residue.

A DNA sequence motif represented as asequence logofor the LexA-binding motif.

Overview

When a sequence motif appears in theexonof agene,it mayencodethe "structural motif"of aprotein;that is a stereotypical element of theoverall structureof the protein. Nevertheless, motifs need not be associated with a distinctivesecondary structure."Noncoding"sequences are nottranslatedinto proteins, andnucleic acidswith such motifs need not deviate from the typical shape (e.g. the "B-form"DNA double helix).

Outside of gene exons, there existregulatory sequencemotifsand motifs within the "junk",such assatellite DNA.Some of these are believed to affect the shape of nucleic acids(see for exampleRNA self-splicing), but this is only sometimes the case. For example, manyDNA binding proteinsthat have affinity for specificDNA binding sitesbind DNA in only its double-helical form. They are able to recognize motifs through contact with the double helix's major or minor groove.

Short coding motifs, which appear to lack secondary structure, include those thatlabelproteins for delivery to particular parts of acell,or mark them forphosphorylation.

Within a sequence ordatabaseof sequences, researchers search and find motifs using computer-based techniques ofsequence analysis,such asBLAST.Such techniques belong to the discipline ofbioinformatics.See alsoconsensus sequence.

Motif Representation

Consider theN-glycosylation site motif mentioned above:

Asn, followed by anything but Pro, followed by either Ser or Thr, followed by anything but Pro

This pattern may be written asN{P}[ST]{P}whereN= Asn,P= Pro,S= Ser,T= Thr;{X}means any amino acid exceptX;and[XY]means eitherXorY.

The notation[XY]does not give any indication of the probability ofXorYoccurring in the pattern. Observed probabilities can be graphically represented usingsequence logos.Sometimes patterns are defined in terms of a probabilistic model such as ahidden Markov model.

Motifs and consensus sequences

The notation[XYZ]meansXorYorZ,but does not indicate the likelihood of any particular match. For this reason, two or more patterns are often associated with a single motif: the defining pattern, and various typical patterns.

For example, the defining sequence for theIQ motifmay be taken to be:

[FILV]Qxxx[RK]Gxxx[RK]xx[FILVWY]

wherexsignifies any amino acid, and the square brackets indicate an alternative (see below for further details about notation).

Usually, however, the first letter isI,and both[RK]choices resolve toR.Since the last choice is so wide, the patternIQxxxRGxxxRis sometimes equated with the IQ motif itself, but a more accurate description would be aconsensus sequencefor the IQ motif.

Pattern description notations

Several notations for describing motifs are in use but most of them are variants of standard notations forregular expressionsand use these conventions:

• there is an Alpha bet of single characters, each denoting a specific amino acid or a set of amino acids;
• a string of characters drawn from the Alpha bet denotes a sequence of the corresponding amino acids;
• any string of characters drawn from the Alpha bet enclosed in square brackets matches any one of the corresponding amino acids; e.g.[abc]matches any of the amino acids represented byaorborc.

The fundamental idea behind all these notations is the matching principle, which assigns a meaning to a sequence of elements of the pattern notation:

a sequence of elements of the pattern notation matches a sequence of amino acids if and only if the latter sequence can be partitioned into subsequences in such a way that each pattern element matches the corresponding subsequence in turn.

Thus the pattern[AB] [CDE] Fmatches the six amino acid sequences corresponding toACF,ADF,AEF,BCF,BDF,andBEF.

Different pattern description notations have other ways of forming pattern elements. One of these notations is the PROSITE notation, described in the following subsection.

PROSITE pattern notation

ThePROSITEnotation uses theIUPACone-letter codes and conforms to the above description with the exception that a concatenation symbol, '-', is used between pattern elements, but it is often dropped between letters of the pattern Alpha bet.

PROSITE allows the following pattern elements in addition to those described previously:

• The lower case letter 'x' can be used as a pattern element to denote any amino acid.
• A string of characters drawn from the Alpha bet and enclosed in braces (curly brackets) denotes any amino acid except for those in the string. For example,{ST}denotes any amino acid other thanSorT.
• If a pattern is restricted to the N-terminal of a sequence, the pattern is prefixed with '<'.
• If a pattern is restricted to the C-terminal of a sequence, the pattern is suffixed with '>'.
• The character '>' can also occur inside a terminating square bracket pattern, so thatS[T>]matches both "ST"and"S>".
• Ifeis a pattern element, andmandnare two decimal integers withm<=n,then:
  • e(m)is equivalent to the repetition ofeexactlymtimes;
  • e(m,n)is equivalent to the repetition ofeexactlyktimes for any integerksatisfying:m<=k<=n.

Some examples:

• x(3)is equivalent tox-x-x.
• x(2,4)matches any sequence that matchesx-xorx-x-xorx-x-x-x.

The signature of the C2H2-typezinc fingerdomain is:

• C-x(2,4)-C-x(3)-[LIVMFYWC]-x(8)-H-x(3,5)-H

Matrices

A matrix of numbers containing scores for each residue or nucleotide at each position of a fixed-length motif. There are two types of weight matrices.

• A position frequency matrix (PFM) records the position-dependent frequency of each residue or nucleotide. PFMs can be experimentally determined from SELEX experiments or computationally discovered by tools such as MEME using hidden Markov models.
• Aposition weight matrix(PWM) contains log odds weights for computing a match score. A cutoff is needed to specify whether an input sequence matches the motif or not. PWMs are calculated from PFMs. PWMs are also known as PSSMs.

An example of a PFM from theTRANSFACdatabase for the transcription factor AP-1:

Pos	A	C	G	T	IUPAC
01	6	2	8	1	R
02	3	5	9	0	S
03	0	0	0	17	T
04	0	0	17	0	G
05	17	0	0	0	A
06	0	16	0	1	C
07	3	2	3	9	T
08	4	7	2	4	N
09	9	6	1	1	M
10	4	3	7	3	N
11	6	3	1	7	W

The first column specifies the position, the second column contains the number of occurrences of A at that position, the third column contains the number of occurrences of C at that position, the fourth column contains the number of occurrences of G at that position, the fifth column contains the number of occurrences of T at that position, and the last column contains the IUPAC notation for that position. Note that the sums of occurrences for A, C, G, and T for each row should be equal because the PFM is derived from aggregating several consensus sequences.

Motif Discovery

Overview

The sequence motif discovery process has been well-developed since the 1990s. In particular, most of the existing motif discovery research focuses on DNA motifs. With the advances in high-throughput sequencing, such motif discovery problems are challenged by both the sequence pattern degeneracy issues and the data-intensive computational scalability issues.

Process of discovery

A flowchart depicting the process of motif discovery

Motif discovery happens in three major phases. A pre-processing stage where sequences are meticulously prepared in assembly and cleaning steps. Assembly involves selecting sequences that contain the desired motif in large quantities, and extraction of unwanted sequences using clustering. Cleaning then ensures the removal of any confounding elements. Next there is the discovery stage. In this phase sequences are represented using consensus strings orPosition-specific Weight Matrices (PWM).After motif representation, an objective function is chosen and a suitable search algorithm is applied to uncover the motifs. Finally the post-processing stage involves evaluating the discovered motifs.

De novomotif discovery

There are software programs which, given multiple input sequences, attempt to identify one or more candidate motifs. One example is theMultiple EM for Motif Elicitation(MEME) algorithm, which generates statistical information for each candidate.There are more than 100 publications detailing motif discovery algorithms; Weirauchet al.evaluated many related algorithms in a 2013 benchmark.Theplanted motif searchis another motif discovery method that is based on combinatorial approach.

Phylogenetic motif discovery

Motifs have also been discovered by taking aphylogeneticapproach and studying similar genes in different species. For example, by aligning the amino acid sequences specified by the GCM (glial cells missing) gene in man, mouse andD. melanogaster,Akiyama and others discovered a pattern which they called theGCM motifin 1996.It spans about 150 amino acid residues, and begins as follows:

WDIND*.*P..*...D.F.*W***.**.IYS**...A.*H*S*WAMRNTNNHN

Here each.signifies a single amino acid or a gap, and each*indicates one member of a closely related family of amino acids. The authors were able to show that the motif has DNA binding activity.

A similar approach is commonly used by modernprotein domaindatabases such asPfam:human curators would select a pool of sequences known to be related and use computer programs to align them and produce the motif profile (Pfam usesHMMs,which can be used to identify other related proteins.A phylogenic approach can also be used to enhance thede novoMEME algorithm, with PhyloGibbs being an example.

De novomotif pair discovery

In 2017, MotifHyades has been developed as a motif discovery tool that can be directly applied to paired sequences.

De novomotif recognition from protein

In 2018, aMarkov random fieldapproach has been proposed to infer DNA motifs fromDNA-binding domainsof proteins.

Motif Discovery Algorithms

Motif discovery algorithms use diverse strategies to uncover patterns in DNA sequences. Integrating enumerative, probabilistic, and nature-inspired approaches, demonstrate their adaptability, with the use of multiple methods proving effective in enhancing identification accuracy.

Enumerative Approach:

Initiating the motif discovery journey, the enumerative approach witnesses algorithms meticulously generating and evaluating potential motifs. Pioneering this domain are Simple Word Enumeration techniques, such as YMF and DREME, which systematically go through the sequence in search of short motifs. Complementing these, Clustering-Based Methods such as CisFinder employ nucleotide substitution matrices for motif clustering, effectively mitigating redundancy. Concurrently, Tree-Based Methods like Weeder and FMotif exploit tree structures, and Graph Theoretic-Based Methods (e.g., WINNOWER) employ graph representations, demonstrating the richness of enumeration strategies.

Probabilistic Approach:

Diverging into the probabilistic realm, this approach capitalizes on probability models to discern motifs within sequences. MEME, a deterministic exemplar, employs Expectation-Maximization for optimizing Position Weight Matrices (PWMs) and unraveling conserved regions in unaligned DNA sequences. Contrasting this, stochastic methodologies like Gibbs Sampling initiate motif discovery with random motif position assignments, iteratively refining the predictions. This probabilistic framework adeptly captures the inherent uncertainty associated with motif discovery.

Advanced Approach:

Evolving further, advanced motif discovery embraces sophisticated techniques, withBayesian modelingtaking center stage. LOGOS and BaMM, exemplifying this cohort, intricately weave Bayesian approaches andMarkov modelsinto their fabric for motif identification. The incorporation of Bayesian clustering methods enhances the probabilistic foundation, providing a holistic framework for pattern recognition in DNA sequences.

Nature-Inspired and Heuristic Algorithms:

A distinct category unfolds, wherein algorithms draw inspiration from the biological realm.Genetic Algorithms (GA),epitomized by FMGA and MDGA,navigate motif search through genetic operators and specialized strategies. Harnessing swarm intelligence principles,Particle Swarm Optimization (PSO),Artificial Bee Colony (ABC)algorithms, andCuckoo Search (CS)algorithms, featured in GAEM, GARP, and MACS, venture into pheromone-based exploration. These algorithms, mirroring nature's adaptability and cooperative dynamics, serve as avant-garde strategies for motif identification. The synthesis of heuristic techniques in hybrid approaches underscores the adaptability of these algorithms in the intricate domain of motif discovery.

This chart shows many different types of algorithms used in the discovery of sequence motifs and their categories

Motif Cases

Three-dimensional chain codes

TheE. colilactoseoperonrepressor LacI (PDB:1lcc​ chain A) andE. colicatabolite gene activator (PDB:3gap​ chain A) both have ahelix-turn-helixmotif, but their amino acid sequences do not show much similarity, as shown in the table below. In 1997, Matsuda,et al.devised a code they called the "three-dimensional chain code" for representing the protein structure as a string of letters. This encoding scheme reveals the similarity between the proteins much more clearly than the amino acid sequence (example from article):The code encodes thetorsion anglesbetween Alpha -carbons of theprotein backbone."W" always corresponds to an Alpha helix.

	3D chain code	Amino acid sequence
1lccA	TWWWWWWWKCLKWWWWWWG	LYDVAEYAGVSYQTVSRVV
3gapA	KWWWWWWGKCFKWWWWWWW	RQEIGQIVGCSRETVGRIL

See also

• Biomolecular structure
• Mammalian Motif Finder
• MochiView
• Multiple EM for Motif Elicitation
• Nucleic acid sequence
• Protein primary structure
• Protein I-sites
• Sequence logo
• Sequence mining
• Structural motif
• Short linear motif
• Conserved sequence
• Protein domain
• Structural motif

References

Further reading

• Kadaveru K, Vyas J, Schiller MR (May 2008)."Viral infection and human disease--insights from minimotifs".Frontiers in Bioscience.13(13): 6455–71.doi:10.2741/3166.PMC2628544.PMID18508672.
• Stormo GD (January 2000). "DNA binding sites: representation and discovery".Bioinformatics.16(1): 16–23.doi:10.1093/bioinformatics/16.1.16.PMID10812473.

Primary sources

• Altarawy D, Ismail MA, Ghanem S (2009). "MProfiler: A Profile-Based Method for DNA Motif Discovery".Pattern Recognition in Bioinformatics.Lecture Notes in Computer Science. Vol. 5780. pp. 13–23.doi:10.1007/978-3-642-04031-3_2.ISBN978-3-642-04030-6.
• Schiller MR (2007). "Minimotif miner: a computational tool to investigate protein function, disease, and genetic diversity".Current Protocols in Protein Science.48(1). Wiley: 2.12.1–2.12.14.doi:10.1002/0471140864.ps0212s48.ISBN978-0471140863.PMID18429315.S2CID10406520.
• Balla S, Thapar V, Verma S, Luong T, Faghri T, Huang CH, et al. (March 2006)."Minimotif Miner: a tool for investigating protein function".Nature Methods.3(3): 175–7.doi:10.1038/nmeth856.PMID16489333.S2CID15571142.