Computational Biology and Language - Ambient Intelligence for Scientific Discovery - page 28

Fig. 2. Analogy between natural language and “protein sequence language”. Words
combine into sentences to convey meaningful content. Processing words in a document
can convey the topic of the document and its meaningful content. Biological sequences
are analogous to the raw texts, processing which would yield higher level information
on the behavioral properties of the physical entities corresponding to the sequences.

in both cases. (ii) Fundamental units of human languages include higher order
structures, paralleled by domains, subunits or functionally linked proteins. (iii)
Computer-based derivation of meaning from text is analogous to the prediction
of structure and function from primary sequence data. Therefore, the analogy
has led to the wide application of methods used in language technologies to the
study of biological sequences [1, 5, 6]. Some examples for the use of linguistic
approaches for bioinformatics can be found in refs. [1, 5-12]. Recently, proba-
bilistic language models have been used to improve protein domain boundary
detection [13], to predict transmembrane helix boundaries [14] and in genome
comparison [15, 16]. Finally, latent semantic analysis, a technique used in text
summarization, has been used for secondary structure prediction [17] and topic
segmentation of text or radio speech. The feature prediction methods of Yule's
Q-statistic [18] and mutual information [19] have been applied to the membrane
protein boundary prediction problem.

2.1 Protein Sequence Language

Like strings of letters and words in a text, protein sequences are linear chains
of amino acids. The amino acid is one of the fundamental building blocks in
protein sequences. This is illustrated schematically in Fig. 3. Each amino acid
has a common component shown in Fig. 3A. In the protein chain, the amino
acids are connected through this component forming the backbone of the pro-
tein. The side chains, represented by R in Fig. 3A, can be one of 20 different