Scientists develop new way to decipher hidden messages in symbols

Sep 27, 2010 By Lisa Zyga feature
A motif network of the human proteome can be used to extract functional protein domains. This network consists of recurrent strings of 3 letters, called 3-motifs. Each node belongs to one of 15 communities, which are labeled with different colors and numbers. Most of the communities can be associated with a functional domain. Image credit: Roberta Sinatra, et al.

(PhysOrg.com) -- Almost all information, in a sense, can be represented by symbols. In order to extract this embedded information, the symbols and the rules governing their sequence formation need to be deciphered. There are many examples of information residing in symbols, although the most familiar is probably written language. In addition to the sequences of letters that make up words, and sequences of words that make up sentences, there are lexical and grammatical rules that govern how letters and words can be combined, respectively, so that not all sequences of letters and words are possible. In a recent study, a group of scientists from Italy has developed a generic method to extract information from any type of symbolic sequential data, even when a "dictionary" of symbol sequences is not known beforehand.

The researchers, Roberta Sinatra, Daniele Condorelli, and Vito Latora, from the University of Catania and the Scuola Superiore di Catania, are publishing their study in an upcoming issue of . As they explained, a few of the many examples where information has to be extracted from sequences of symbols are , DNA nucleotides, musical notation, dance movements, texts written in an unknown language, and others. Having a general method to extract the information in any type of symbol sequence could be extremely useful for deciphering encoded data.

"I think that it is interesting that one can construct a lexicon for every non-random collection of symbolic data, just by means of ," Sinatra told PhysOrg.com. "In fact, if a sequence of symbols encodes some information, it cannot be random and probably is made up of fundamental units that perform the same role that words do in language. Therefore, by extracting these significant units, it is possible to construct a dictionary also for proteins, dances, and music, since they can all be represented in terms of sequences of symbols that show non-trivial statistical properties."

In their method, the researchers first converted the symbol sequences into a based on a dictionary of significant strings of symbols extracted from the sequences. They called these significant strings “motifs,” which are the equivalent of words in a language, since they deviate from randomness. The motifs represent the fundamental “bricks” that sequences are made of by following rules of combination syntax. When converted to a network, these motifs form the nodes of the network. The links between the nodes represent a significant occurrence of two motifs in the same sequence (the equivalent of a phrase or sentence). So a weighted, directed link between two nodes means that the nodes often appear together in a sequence in a certain order. For example, if “the” and “end” were two nodes, there would likely be a directed link between them to represent the existence of the common phrase “the end.”

“We know that if we type at random on a keyboard, it is unlikely that we end up forming a sentence or even a word that makes sense,” Sinatra explained. “Similarly, we know that if we select some letters, say E,H,M,O, we know that we need to put them in a specific order, for example HOME and not HMOE, to encode some information in it. By looking at how many times the sequence HOME and the sequence HMOE appear in a text, we can understand which string contains a message and which is just there by chance (due to a typo, for example). However, we know that important information is contained not only at the level of words, but also in sentences and in general in how words are coexpressed. So it is unlikely that we find the words "the" and "for" next to each other or a verb followed by another verb ("I sit go"). This is why we introduce the concept of the network of motifs: it embeds the information of how words correlate in sequences of symbols.”

By analyzing the network of motifs, the researchers could identify significant patterns, and then extract important information encoded in the original data simply from the way the network is structured. In particular, information about the network's community structures, i.e., the groups of nodes that are tightly connected among themselves and weakly connected to the rest of the network, proved helpful for extracting encoded messages. The researchers demonstrated the approach for three different data sets: the human proteome (the protein equivalent of the genome), Twitter posts, and dynamical systems. For instance, in the proteome example, the communities of the motif network can identify those parts of proteins (sequences of amino acids) called functional domains that specify the protein function. In these systems and others, the motif network approach can be very useful for processing information from extremely large amounts of symbolic data.

“With this method, we are able to compact information deriving from an entire ensemble of sequences, like all the proteins of a species or all the tweets posted in one day in Twitter, in just one object: the network of motifs,” Sinatra said. “Of course, one could study in great detail only one sequence, for example one protein or one post, and have complete knowledge of what information that protein or post means. But this would be equivalent to reading just one sentence of an entire book: what can one understand from just one sentence if a book is made up of thousands of them? This is why we usually read the summary on the back cover of a book. Well, the network of motifs plays exactly this role: it 'summarizes' the entire ensemble of sequences, providing information on what the main message is.”

Explore further: New microscope collects dynamic images of the molecules that animate life

More information: Roberta Sinatra, Daniele Condorelli, and Vito Latora. “Networks of motifs from sequences of symbols.” To be published. Available at arXiv:1002.0668v2. arxiv4.library.cornell.edu/abs/1002.0668v2

4.7 /5 (41 votes)

Related Stories

Learning the language of gene expression

Jan 19, 2007

Researchers have taken a major step towards understanding the language of gene regulation in the fruitfly Drosophila and they expect the technique to be rapidly applicable to understanding the effects of genome variation ...

Single neurons can detect sequences

Aug 12, 2010

Single neurons in the brain are surprisingly good at distinguishing different sequences of incoming information according to new research by UCL neuroscientists.

Recommended for you

Cooling with molecules

Oct 22, 2014

An international team of scientists have become the first ever researchers to successfully reach temperatures below minus 272.15 degrees Celsius – only just above absolute zero – using magnetic molecules. ...

User comments : 6

Adjust slider to filter visible comments by rank

Display comments: newest first

banjopete
3 / 5 (2) Sep 27, 2010
Someone should put this to use deciphering Linear A script.
kasen
4.3 / 5 (6) Sep 27, 2010
Voynich manuscript, anyone?
Macksb
3.3 / 5 (3) Sep 27, 2010
There are two related network patents that should be applied to the intellectual framework described in this article. US Patent 5 734 580 (Rakov)uses perfect difference sets and number theory as the basis for a network in which every node is connected to every other node in the most efficient manner. It is perfectly symmetric as well. Diameter of 2. Easily scalable. It's described in various articles by Parhami, as a "perfect difference network." The second patent, US 6 330 706 (Rakov et al.) is for a method of adding logic to the first patent, to create an intelligent network. Briefly, the logic system involves weights on the interconnecting lines and functions at intermediate nodes. Nodes that are connected more often could have more weight on the connecting lines, for example, and any other parameters of interest to the researchers could be specified. The perfect symmetry of the PDN network (first patent) makes it possible to add such a powerful logic system.
Macksb
3.5 / 5 (4) Sep 28, 2010
Continuing with my prior post, take a look at the network diagram at the top of the article. The network is fairly complex and disjointed, even though the number of nodes is modest--maybe 70 to 100 nodes in all. All of the information in the pictured network could be incorporated in a far more elegant and tractable package in a PDN network. The logical connections could be specified in a more quantitative and subtle way, with finer distinctions. More importantly, the PDN network and logic system can be scaled much higher, to millions of nodes and beyond. Ordinary network designs become impossibly complex when the number of nodes reaches 500 or 1000, especially if one has to specify varying logical connections within the network. Try taking that simple picture and using it to show nodes in protein sequences or DNA and their connections.

The article describes some impressive work with broad potential. I hope the authors and others will try painting it on the PDN network canvas.
Antonio_Zippo
5 / 5 (2) Sep 28, 2010
In 90's the variable order Markov models were introduced. They outperform the K-order Markov Models (called k-motifs in the work) and require the minimum amount of space. A lot of independent authors showed their better performances (Mohammed Zaki, Begleiter et al). Moreover the extraction of a graph structure starting from the estimated models it's been already proposed by Rudi Cilibrasi et al in the last decade.
Eikka
not rated yet Oct 08, 2010
This is why statistical machine translation is doomed to fail.

what if "hmoe" is just a rarely used word instead of a typo?