Microsoft Word - Instaseq manuscript.doc

Searching the World-Wide-Web using nucleotide and
peptide sequences

Natarajan Ganesan

, Nicholas F. Bennett, Bala Kalyanasundaram, Mahe Velauthapillai,

and Richard Squier

Department of Computer Science, Bioinformatics & Computational Biosciences Unit,

329A Saint Mary's Hall Georgetown University 37th and O Streets, NW Washington,

DC 20057-1232

Keywords: World-Wide-Web, search, nucleotide or peptide sequences

Corresponding author: Natarajan Ganesan,

ng6@georgetown.edu

Abstract

Background - No approaches have yet been developed to allow instant searching of the

World-Wide-Web by just entering a string of sequence data. Though general search

engines can be tuned to accept `processed' queries, the burden of preparing such `search

strings' simply defeats the purpose of quickly locating highly relevant information.

Unlike `sequence similarity' searches that employ dedicated algorithms (like BLAST)

to compare an input sequence from defined databases, a direct `sequence based' search

simply locates quick and relevant information about a blunt piece of nucleotide or

peptide sequence. This approach is particularly invaluable to all biomedical researchers

who would often like to enter a sequence and quickly locate any pertinent information

before proceeding to carry out detailed sequence alignment.

Results - Here, we describe the theory and implementation of a web-based front-end

for a search engine, like Google, which accepts sequence fragments and interactively

retrieves a collection of highly relevant links and documents, in real-time. e.g. flat files

like patent records, privately hosted sequence documents and regular databases.

Conclusions - The importance of this simple yet highly relevant tool will be evident

when with a little bit of tweaking, the tool can be engineered to carry out searches on all

kinds of hosted documents in the World-Wide-Web.

Availability - Instaseq is free web based service that can be accessed by visiting the

following hyperlink on the WWW

http://instaseq.georgetown.edu

Background

The amount of hosted sequence-data in biomedical sciences continues to double

rapidly, and much of this information is accessible on the Web. There exists a wide

variety of tools in the information pyramid (Figure 1), at specialized and sub-

specialized levels for sequence analysis and manipulation e.g. BLAST, structure

prediction servers

1-3

, and profiling tools

4-6

. These tools operate just above the ground-

level mosaic of sequence records and other raw data present mainly in databases and

other resources. At a higher tier are search engines like Entrez

, Sequerome

and

Bioinformatic Harvester

which provide information collected from a variety of

databases. However, these engines confine their queries to a relatively small number of

specific sources, and do not access the large amounts of data found in the Web, such as

text documents and newly created data files in a variety of plain, flat-file formats hosted

by special-interest groups. General search engines do access the Web at large, and if

properly used, users can enter their sequences directly to locate relevant records.

Unfortunately, such searches usually fail to produce the desired results because the user

does not adequately prepare the search string. The burden of preparing the search string

simply defeats the purpose of quickly locating highly relevant information.

The following example demonstrates the problem. Consider the single-line peptide

sequence below,

ANLGCWMLVLFVATWSDLGLCKKRPKPGGWNTGGSRYPG

A user, unaware of the existence of its NCBI record, might try to grab any relevant

information by pasting the sequence into a

GOOGLE

TM search window. This would not

yield any information when, in fact, there exist many records on the web containing this

sequence. Queries based directly on

FASTA format

rarely yield results. Alternately, the

user could use BLAST

to perform a sequence alignment, but this can be time

consuming and requires careful attention to correctly adjust a variety of matrix and

relevance parameters. Since neither of these methods obtains adequate results quickly,

there exists a need for a simple and fast search tool to locate relevant information, in

contrast to elaborate alignment and profiling tools. Despite its limitations, the sheer

power and speed of Google search strongly appeals to end-users who would simply

require a quick and crude method to locate bare minimum piece of information about an

unknown sequence. One can then search for further references and perform specialized

searches and detailed alignment analyses.

Implementation

One way to address the problem in the above example would be to insert a space

into the sequence at every tenth character and submit this re-formatted sequence as a

quoted query string to the search engine. In this case, our search retrieved records for a

Prion protein with sources ranging from databases to shared flat-files

. What happens is

that each ten-letter subsequence is treated as a keyword by the Google, and it searches

for exact matches to the entire phrase. Since many sequence records on the Web

traditionally use this format, this approach yields useful results. However, the user's

ten-letter formatted sequence could be frame-shifted with respect to the data source.

This can be caused by the user starting with a truncated sequence or the data source

having a record with gaps, for instance. Hence another solution would be to generate

search strings using all ten possible frame shifts, which would greatly improve the

search results. In our experience this is overkill; to assure a minimal, yet useful, search

result containing an adequate number of references, it is sufficient to form search

phrases by taking any two consecutive ten-character keywords from each of the ten

frame shifts, then combining these phrases with the Boolean `OR' operator. More often

than not, such a search retrieves records describing the original source document along

with other highly relevant information, if they exist. The reason for this specificity is

that the probability of another sequence matching twenty characters as two consecutive

ten-letter `words' is extremely low. Even if the search hit is not specific, the matched

sequence is more likely to bear a good `e-value'

with respect to the input sequence.

This simple idea is the basis behind the development and implementation of InstaSeq

(Figure 2).

Results and Discussion

In our experiments, the retrieved documents were found to be well focused and

often pointed to information derived from the input fragment. Databases covered

included ExPASy, Uniprot, HPRD, FreshPatents, and text-format documents which

included many shared flat-files from small sources. For example, pasting the following

fragment of a protein sequence

into InstaSeq,

YLVTHLMGADLNNIVKCQKLTDDHVQFLIYQILRGLKYIHSADIIHRDLKPSNLAVNE

returned this link, among others,

http://pkr.sdsc.edu/html/3D/text/1p38/1p38_snap.html

which is a snapshot of a 3D structure of the "mitogen activated protein kinase"

molecule from Rattus norvegicus.

Our current tool and approach represents only the tip of an iceberg of many

possible extensions that leverage search engine functionality. One goal is to reduce the

likelihood of failing to find relevant information when it exists. For instance, InstaSeq's

query string generation can be easily modified to also search for sequence data hosted in

non-traditional forms such as protein sequences in three-letter format and DNA primer-

sequences hosted by biotechnology companies. Also, nucleotide sequences could easily

be translated into protein sequences using all three shift frames. A separate goal is to

provide comprehensiveness in query results where relevant data does exist. One way of

increasing the coverage of returned information is to use specific keywords directly

related to the input sequence. Dedicated Web crawlers could search for documents

containing sequence information and collect keywords found there. These keywords

could then be indexed to their corresponding sequences in a database, and retrieved

when a search for the sequence has been done. Figure 3 shows a schematic approach of

the method, with keywords being fed to a search engine to return a more comprehensive

list of documents. Providing navigation aids for reviewing search results would be

another opportunity for further development, given that specificity and simplicity are in

competition with comprehensiveness.

Since the results from a Google query can vary significantly due to their

patented page-ranking technology

, InstaSeq may retrieve different document sets at

different times for the same input sequence. However, we do not consider this to be a

significant shortcoming as Web content itself is highly dynamic. Despite its

shortcomings, when compared to a direct Google search

on a batch of test sequences,

InstaSeq yielded more hits which were all highly relevant. The search worked well in all

tested browsers running under a variety of operating systems. InstaSeq is a publicly

available, free, web-based service, and can be accessed via its homepage

http://bioinformatics.georgetown.edu/InstaSeq.htm

Availability and requirements

: InstaSeq can be accessed freely and directly as a

web-server at its homepage -

http://bioinformatics.georgetown.edu/InstaSeq.htm

Acknowledgements:

N.G. conceived and designed the project, and together with N.F.B coded the web based

tool. M.V. and B.K. provided valuable inputs; N.G. and R.S co-wrote the paper.

Correspondence and requests for materials should be addressed to N.G.

(

ng6@georgeotwn.edu

References

1. Altschul,S.F., Gish,W., Miller,W., Myers,E.W. & Lipman,D.J. Basic local

alignment search tool. J. Mol. Biol. 215, 403-410 (1990).

2. Zuker,M. Mfold web server for nucleic acid folding and hybridization prediction.

Nucleic Acids Res. 31, 3406-3415 (2003).

3. Schultz,J., Milpetz,F., Bork,P. & Ponting,C.P. SMART, a simple modular

architecture research tool: identification of signaling domains. Proc. Natl. Acad.

Sci. U. S. A 95, 5857-5864 (1998).

4. Liebel,U., Kindler,B. & Pepperkok,R. 'Harvester': a fast meta search engine of

human protein resources. Bioinformatics. 20, 1962-1963 (2004).

5. Ganesan,N. et al. A web-based interface facilitating sequence to structure analysis

of BLAST alignment reports. Biotechniques . 8-1-2005.

Ref Type: In Press

6. Stothard,P. The sequence manipulation suite: JavaScript programs for analyzing

and formatting protein and DNA sequences. Biotechniques 28, 1102, 1104 (2000).

7. Maglott,D., Ostell,J., Pruitt,K.D. & Tatusova,T. Entrez Gene: gene-centered

information at NCBI. Nucleic Acids Res. 33, D54-D58 (2005).

8. Ganesan,N. InstaSeq Sample Searches.

http://bioinformatics.georgetown.edu/InstaSeq_sample_searches.htm

. 6-15-2005.

Ref Type: Electronic Citation

Supporting material The homepage accepts queries into the search box. Queries are best taken in

FASTA format with length preferably > 30 characters. Upon submitting the query to InstaSeq, a small

javascript on the homepage parses the input sequence into strings of ten-characters in all different frames,

truncates at a particular length , takes the beginning two-words in each frame and finally combines all

these generated `words' with an `OR' Boolean operator. This prepared string is then submitted to Google

search (free source code) which then returns the records from the web. The default is a web search but

users can also select a

SEQUEROME

http://sequerome.georgetown.edu

search, to perform a detailed

BLAST profiling analysis. The file-types listed include documents enlisted as .doc, .pdf and .ppt thus

allowing access to records stored in all formats. Since Google uses page-ranking technology to pull out

documents from the web, the results of a same query can vary over time. The following link -

http://bioinformatics.georgetown.edu/InstaSeq_sample_searches.htm

- provides performance reports on

the InstaSeq searches. Though the search yields all relevant links pointing to the input sequence, it is not

designed to perform sequence alignment and pattern matching to return records that are similar to the

input sequence, which is best carried out by tools like BLAST and its family of specialized tools. The

search worked well in the browsers tested (MS Internet Explorer, Firefox, and Netscape Navigator) on

several operating systems (Windows, Mac OS X, and Linux). InstaSeq is a publicly available, free, web-

based service, and can be accessed via its homepage -

http://bioinformatics.georgetown.edu/InstaSeq.htm

Legends to Figures

Figure 1 shows the pyramid of gene/protein sequence data hosted on the

World-Wide-Web and the existing approaches and set of informatic tools to deal

with this specialized kind of information.

Figure 2 shows a brief schema behind the working of InstaSeq search tool.

Queries are best taken in FASTA format with length preferably between > 30

characters. Upon submitting the query to InstaSeq, a small javascript on the

homepage parses the input sequence into strings of 10 in all different frames,

truncates at a particular length , take the beginning two-words in each frame

and finally combines all these generated `words' with an `OR' Boolean operator.

This prepared string in then submitted to Google search (free source code)

which then returns the records from the web.

Figure 3 shows the schema of a possible approach by InstaSeq based

searches to cover a wide variety of gene/protein sequence related flat-files

hosted in the World- Wide Web. The main step lies creating mirrored in-house

databases as flat-files and getting them crawled by dedicated crawlers OR

WWW engines. After the user query locates the relevant flat-file, a suitable

program would then generate a brief list of highly relevant keywords which

would be then submitted to Google. This would result in a greatly enhanced set

of relevant documents and links pertaining to the input gene/protein sequence.