ajb at s-crim1.dl.ac.uk ajb at s-crim1.dl.ac.uk
Wed Jan 5 21:43:21 EST 1994

The MOWSE Peptide-Mass Fingerprint server has been updated to show
partial cleavage fragments (marked by `*' in the output). The
format of the output has also changed slightly in order to reduce the
size of the resulting email. Specifically, the `no match' fragments are
reported as a block rather than individually.

A new partials option (Pfactor) is added to down-weight the effect of
partial matches. If the pfactor is not specified then sensible defaults
are taken. See the appended help file for details.


Alan Bleasby
SERC Daresbury Laboratory
Warrington WA4 4AD

		The MOWSE peptide mass database:

		Imperial Cancer Research Fund


		SERC Daresbury Laboratory

        	D.J.C. Pappin, P. Hojrup and A.J. Bleasby
		'Rapid Identification of Proteins by
		Peptide-Mass Fingerprinting'.
       	 	Current Biology (1993), vol 3, 327-332.

		InterNet server version:

Table of Contents:

	[1] Background.

	[2] Construction of the MOWSE database.

		[2.1] Source database.
		[2.2] Calculation of Molecular weight fragments.

	[3] Running database searches via e_mail.

	[4] Example of mail query format.
	[5] Results listing.

	[6] Database structure.

		[6.1] MOWSE database structure.
		[6.2] The MW primary fragment molecular weight file.
		[6.3] The MDX file OWL entry index.
		[6.4] The SMW whole sequence molecular weight file.
		[6.5] Program Requirements.
		[6.6] MOWSE Scoring scheme.
		[6.7] Simulation studies.

	[7] General references.

[1] Background:

	Determination of molecular weight has always been an 
important aspect of the characterization of biological molecules. 
Protein molecular weight data, historically obtained by SDS gel 
electrophoresis or gel permeation chromatography, has been used 
establish purity, detect post-translational modification (such as 
phosphorylation or glycosylation) and aid identification. Until 
just over a decade ago, mass spectrometric techniques were typically 
limited to relatively small biomolecules, as proteins and nucleic 
acids were too large and fragile to withstand the harsh physical 
processes required to induce ionization. This began to change with 
the development of 'soft' ionization methods such as fast atom 
bombardment (FAB)[1], electrospray ionisation (ESI) [2,3] and 
matrix-assisted laser desorption ionisation (MALDI)[4], which can 
effect the efficient transition of large macromolecules from 
solution or solid crystalline state into intact, naked molecular 
ions in the gas phase. As an added bonus to the protein chemist, 
sample handling requirements are minimal and the amounts required 
for MS analysis are in the same range, or less, than existing 
analytical methods.
	As well as providing accurate mass information for intact 
proteins, such techniques have been routinely used to produce 
accurate peptide molecular weight 'fingerprint' maps following 
digestion of known proteins with specific proteases. Such maps 
have been used to confirm protein sequences (allowing the 
detection of errors of translation, mutation or insertion), 
characterise post-translational modifications or processing events 
and assign disulphide bonds [5,6]. 
	Less well appreciated, however, is the extent to which such 
peptide mass information can provide a 'fingerprint' signature 
sufficiently discriminating to allow for the unique and rapid 
identification of unknown sample proteins, independent of other 
analytical methods such as protein sequence analysis. 
	The following text describes the construction and use 
of the MOWSE peptide mass database (for MOlecular Weight SEarch) 
at the SERC Daresbury Laboratory. Practical experience has shown 
that sample proteins can be uniquely identified using as few as 3-
4 experimentally determined peptide masses when screened against a 
fragment database derived from over 50,000 proteins. Experimental 
errors of a few Daltons are tolerated by the scoring algorithms, 
permitting the use of inexpensive time-of-flight mass 
spectrometers. As with other types of physical data, such as amino 
acid composition or linear sequence, peptide masses can clearly 
provide a set of determinants sufficiently unique to identify or 
match unknown sample proteins. Peptide mass fingerprints can prove 
as discriminating as linear peptide sequence, but can be obtained 
in a fraction of the time using less material. In many cases, this 
allows for a rapid identification of a sample protein before 
committing to protein sequence analysis. Fragment masses also 
provide structural information, at the protein level, fully 
complementary to large-scale DNA sequencing or mapping projects 

[2] Construction of the MOWSE database.

[2.1] Source database.

	MOWSE was created from the OWL non-redundant composite 
protein sequence database [10,11]. The first InterNet release (version 
20.1) contains some 61,000 protein entries, generating approximately
15,000,000 peptide fragments. The MOWSE fragment database will be updated
with each new release of the parent OWL database (every 2 months or so).
[2.2] Calculation of Molecular weight fragments.

	For each entry in the source OWL database, MOWSE derives both 
whole sequence molecular weight and calculated peptide molecular 
weights for complete digests using the range of cleavage reagents 
and rules detailed in Table 1. Cleavage is disallowed if the 
target residue is followed by proline (except for CNBr or Asp N). 
Glu C (S. aureus V8 protease) cleavages are also inhibited if the 
adjacent residue is glutamic acid.  Peptide mass calculations are 
based entirely on the linear sequence and use the average isotopic 
masses of amide-bonded amino acid residues (IUPAC 1987 relative 
atomic masses). To allow for N-terminal hydrogen and C-terminal 
hydroxyl the final calculated molecular weight of a peptide of N 
residues is given by the equation:

	/  Residue mass + 18.0153

	Molecular weights are rounded to the nearest integer value 
before being entered into the database. Cysteine residues are 
calculated as the free thiol, anticipating that samples are 
reduced prior to mass analysis. CNBr fragments are calculated as 
the homoserine lactone form. Information relating to post-
translational modification (phosphorylation, glycosylation etc.) 
is not incorporated into calculation of peptide masses.
Reagent no.	Reagent			Cleavage rule	
	1	Trypsin			C-term to K/R
	2	Lys-C			C-term to K
	3	Arg-C			C-term to R
	4	Asp-N			N-term to D
	5	V8-bicarb		C-term to E
	6	V8-phosph		C-term to E/D
	7	Chymotrypsin		C-term to F/W/Y/L/M
	8	CNBr			C-term to M

	Table 1: Cleavage reagents modelled by MOWSE.

	Current versions of the MOWSE database also incorporate
calculated peptide Mw's resulting from incomplete or partial cleavages.
At present, this is achieved by computing all nearest-neighbour pairs
for each enzyme or reagent detailed in table 1.

[3] Running database searches by e_mail:

Search queries should be mailed to mowse at daresbury.ac.uk (short form
mowse at dl.ac.uk). Search results will be returned directly to your
e_mail address. Comments, please, to mbdpn at s-crim1.dl.ac.uk.

The  'subject'  field  of  your  email  message is irrelevant - all
parameters must be specified in the body of the message.  The relevant
syntax is given below. Some lines are compulsory, others are optional
(see the  description  of parameters section).
All text is case-insensitive, and MOWSE expects integer data. Non-exponential
floating point syntax is acceptable, but MOWSE will round the data to the 
nearest integer. Whitespace is ignored in an intuitive way.

MOWSE  recognises  the  following  command  lines  which  are  further
described below

The order of lines is irrelevant with the exception of 'begin' and the
'datastart/dataend' commands (see below).
If multiple instances of a command occur then only the FIRST instance
will be recognised

  Every search query MUST start with a 'begin' line. There should only
  be one 'begin' line and all other commands & data should immediately

  Every search query MUST specify a 'reagent' line. The word 'reagent'
  must be followed by one of the supported cleavage reagents. These are:


  A typical reagent line is therefore of the form:

     reagent trypsin

  This line is optional. The supplied number specifies the error
  allowed for mass accuracy of experimental mass determination. If no
  figure is  specified, a default tolerance of 2 Daltons will
  be assumed. If you wish to specify a different tolerance then follow
  the word 'tolerance' with the required number of Daltons e.g.

     tolerance 1

  In this case, supplied peptide masses will be matched to +/- 1
  Daltons. Values of 2-4 are suggested for data obtained by laser-
  desorption TOF instruments. Accuracies of +/- 2 Daltons or better are
  generally only possible using an appropriate internal standard (e.g.
  oxidised insulin B chain) with TOF instruments.
  For electrospray or FAB data, a value of 1 can be selected in most 
  cases. If you have real confidence in mass determination, specify '0' 
  (zero) to limit matches to the nearest integer value (effectively +/- 0.5 
  Daltons). Discrimination is significantly improved by the selection of a 
  small error tolerance.

  This optional line allows you to give the molwt of the whole protein (if
  known). This allows you to limit the search to proteins of this molwt
  plus/minus a 'limit' (see below).
  If unspecified, a whole protein molwt of 0 is assumed which MOWSE
  interprets as "search the whole database". This will include all proteins
  up to the maximum size of just under 700,000 Daltons.
  You can specify any molwt in Daltons with this command e.g.

     SeqMW 90000

  This optional line is used in conjunction with the SeqMW command and
  is meaningless without it. It specifies a percentage. Only proteins
  of the given SeqMW +/- this percentage will be searched. If a SeqMW
  is specified but Filter is unspecified then Filter will default to
  25%. To specify a percentage of 30% use:

     Filter 30

  In this case, a molecular weight of 90,000 Daltons was
  specified and the selection of 30 for the filter restricts the
  search to those proteins with masses from 63,000 to 117,000
  Daltons. A value of 25 is suggested for initial searches, which
  can be progressively widened for subsequent search attempts if no
  matches are found. Discrimination is best when the filter
  percentage is narrow, but some Mw estimates (particularly from SDS
  gels) should be given considerable  allowance for error.

  This specifies the weighting given to partially-cleaved peptide
  fragments, with a range from 0.1 to 1.0. If not specified, default
  values are set to 0.2 for whole database searches (where SeqMW is set
  to zero) or 0.4 for limited searches (where SeqMW and Filter are
  specified). The factor effectively down-weights the score awarded to
  a partial fragment by the specified amount. For example, a Pfactor
  of 0.25 will reduce the score of partial fragments to 25% (one quarter)
  of the score of a complete ('perfect') peptide cleavage fragment of
  equal mass.

  Computing all possible nearest-neighbour partial fragments adds
  significantly to the number of peptides entered in the database
  (by a factor of two). The major effect of this is to increase
  the background score by increasing the number of random Mw matches,
  which can significantly reduce discrimination. The use of a low 
  Pfactor (eg 0.1 - 0.3) is a useful way of limiting this effect - 
  partial peptide matches will add a little to the cumulative frequency 
  score, but without compromising discrimination.

  More experienced users can utilise the Pfactor to optimize searches
  where the peptide Mw data contain a significant proportion of
  partial cleavage fragments (eg > 30%). In such cases, setting the Pfactor 
  within the range 0.4 - 0.6 can help to improve discrimination.
  Conversely, if the digestion is perfect, with no partial fragments
  present, the lowest Pfactor of 0.1 will give maximum discrimination.

  If this line appears anywhere in the text body then only this
  document will be emailed to you!

Specifying molecular weight data:

  This line is compulsory and must mark the beginning of the fragment
  molecular weight data. The molecular weights must follow this line,
  each molecular weight being on a line of its own. Masses (M not M[H+])
  are accepted in any order (ascending,descending or mixed).
  Peptide masses can be entered as integers or floating-point values,
  the latter being rounded to the nearest integer value for the search.
  It is suggested that peptide masses should be selected from
  the range 700-4000 Daltons. This range balances the fact that very
  small peptides give little discrimination and minimizes the
  frequency of partially-cleaved peptides.

  As a general rule, users are advised to identify and remove peptide
  masses resulting from autodigestion of the cleavage enzyme (e.g tryptic
  fragments of trypsin), best obtained by MS analysis of control digests
  containing only the enzyme.

  This line marks the end of the fragment data and must be on a line
  of its own. If the data block is at the end of the mail query then
  this line is optional.

[4] Example mail query:
The following example text should form the body of a typical e_mail query:

Reagent 	Trypsin
Tolerance	2
SeqMW		90000
Filter		30
Pfactor		0.4

NOTE: The described structure will allow for multiple search requests
per e_mail message. This feature is NOT presently supported and users
are asked to submit separate e_mail messages for each search request.

[5] Results listing (InterNet version).

        The MOWSE search program outputs a listing file
  containing the following information.

Specified search parameters.

        Includes all specified parameters such as digest reagent, 
  specified error tolerance, specified intact protein Mw and Mw filter
  percentage. All supplied peptide Mws are listed in descending order, 
  followed by the total number of entries scanned during the search.

Short 'hit' listing.

        The top 30 scoring proteins are then listed in descending
  order, details including the OWL entry code and brief text
 identifiers. Details are limited to the top 30 scores as a deliberate
  compromise to keep the result listings as short as possible for
  e_mail return.

Detailed 'hit' listing.

        The top 30 entries are then listed in more detail.The first
  line includes the OWL entry code, the MOWSE search score
  (typically a few powers of 10), the 'hit' protein Mw and finally
  an 'accuracy' ratio calculated by dividing 'hits' by the total
  number of peptides used for the search. This cannot be used to
  ascribe significance, but experience has shown that anything below
  0.3 is generally not worth pursuing. Line 2 is the OWL
  text identifier. Subsequent lines list 'hit' and 'miss' peptides,
  with the appropriate start, end and corresponding sequences of correct
  peptide matches.

  	Matching peptides marked with a '*' denote partially-cleaved

  OWL is available via anonymous ftp from

[6] Database structure.
[6.1] MOWSE database structure.

The database consists of three binary files:

i) MOWSE.MW 		The primary file containing the fragment 
			molecular weights.

ii) MOWSE.MDX 		Index file relating OWL identifier codes 
			to the molecular weight information 
			in the primary Mw file.

iii) MOWSE.SMW		Calculated molecular weights of intact OWL 

	The query program accesses the binary information 
transparently from the user viewpoint. In the internal 
representation the molecular weight (and other) integers are 
stored as 4-byte machine specific quantities. The binary files can 
be transferred between machines of the same 'endian' nature, but 
'cross-endian' transfer is not possible. The MOWSE software allows 
recreation of the files on any platform supporting a standard C 
language compiler. The organisation of the database files is 
described below.

[6.2] The MW primary fragment molecular weight file.

	Fragment molecular weight entries in this file map 
sequentially to the order of entries within the source (OWL) 
protein sequence file.  Each MW file entry consists of 4 blocks 
and are shown below. The MW entries are catenated.

	Block 1	OWL Entry Code	20 bytes
	Block 2	OWL Title Line	80 bytes
	Block 3	Reagent Table	80 bytes
	Block 4	Reagent 1	4 byte
		Reagent 2 	
		Reagent 3	
		Reagent 8	

	The OWL  entry code is the unique identifier of the source 
protein sequence within the OWL database. The code is padded to 20 
bytes using null characters. The title line contains the 
descriptive text of the source protein sequence as given in OWL 
(null terminated). When the OWL title line is longer than 80 
characters, the MW entry is truncated.
	Block 3 is a table of 20 consecutive 4-byte integers, one for 
each allowed reagent in the current MOWSE implementation. The 
order of the integers follows the order given in Table 1. Each 
integer holds the number of fragments derived from the target 
sequence using the associated enzyme (e.g. the fourth integer 
represents the number of fragments produced by the theoretical 
complete digest of a sequence with Arg-C). Unused slots (9-20) are 
assigned zero value.

	The  final block consists of 4-byte integers holding the 
fragment molecular weights. Again, the lists mirror the order of 
the enzymes given in Table 1. Fragment peptides for each enzyme 
are sorted on the basis of continually decreasing calculated 
molecular weight.

[6.3] The MDX file OWL entry index.

The MDX file consists of three blocks as shown below:

	Block 1		Header Information
	Block 2		FTELL entries
	Block 3 	Buckets

HEADER: This contains six 4-byte binary integers containing

         	a) The number of OWL database entries
         	b) The maximum length of a unique identifier code
         	c) The number of buckets (see text)
         	d) The start position of Block 2 within the MDX file.
         	e) Reserved
         	f) The start position of Block 3 within the MDX file


	This is a block of 4-byte binary integers giving the start 
position of MW file entries. The list reflects the sequential 
order of the entries in the MW file. The positions are the 
displacement, in bytes, of the start of an entry from the 
beginning of the MW file.


	Given a unique OWL identifier, a hashing algorithm is used to 
quickly locate fragment molecular weights. The initial twelve 
characters of the identifier are grouped into consecutive binary 
pairs thereby yielding six 2-byte integers. These are hashed using 
the equation:

T = (int4 xor int2) * 26 + (int1 xor int3) * 23 + (int0 xor int5)

Identifiers having less than twelve characters are padded out with 
spaces. The final hash value 'B' is given by:

	B = T modulo PRIME

where the value B represents a 'bucket' into which the identifier 
will slot. The number 'PRIME' is a prime number giving the number 
of  buckets within the file and is the third integer in the header 
block. Each bucket is 512 bytes long and contains a maximum of 
thirty two 16-bit entries.  Each bucket entry consists of the 
twelve bytes of unique OWL identifier code (truncated or padded 
with spaces) followed by a 4-byte integer. The integer is the  
sequential  position of the MOWSE entry (the first entry is 
represented by the value 1). The MDX indexing software chooses the 
prime number value such that there is no bucket containing more 
than the allowed number of entries.

	In order to locate the start position of an entry in the MW 
file the entry code is hashed to determine the bucket number. The 
relevant bucket is searched to find the sequential position of the 
entry. The sequential position is then used as an index into Block 
2 to find the displacement of the MW entry with respect to the 
start of the file.

[6.4] The SMW whole sequence molecular weight file.

	Each sequence molecular weight is represented as a 4-byte 
integer. The molecular weights are stored in the same order as the 
entries for each protein in the primary fragment MW file.

[6.5] Program Requirements.

	The MOWSE search program accepts a single text file 
containing a list of experimentally-determined masses, generally 
selected from the range 700-4,000 Daltons to reduce the influence 
of partial cleavage products. The program outputs a ranked hit 
list comprising the top 30 scores, with information including the 
OWL entry name, text identifiers, final accumulated scores, matching
peptide sequences and hit versus miss tallies. User-selectable search
parameters include an error tolerance (default +/- 2 Daltons), selection
of the enzyme or reagent used and an intact protein Mw (optional, if 
	For each peptide Mw entry in the data file, MOWSE matches 
individual fragment molecular weights (FMWs) with database entry 
molecular weights (DBMWs). A 'hit' is scored when the following 
criterion is met:

	DBMW-tolerance-1 < FMW < DBMW+tolerance+1

If an intact protein Mw is specified (SMW) then the program 
prompts for a molecular weight filter percentage (MWFP). MOWSE 
then restricts the search to those entries which match the 
following criteria:

	R = SMW x MWFP / 100
   	0 < SMW-R < MOWSE entry Mol.wt. < SMW+R

	Default search parameters are a tolerance of +/- 2 Daltons, 
intact Mw specified and the MWFP set to 25.

[6.6] MOWSE Scoring scheme.

	The final scoring scheme is based on the frequency of a 
fragment molecular weight being found in a protein of a given 
range of molecular weight. OWL database sequence entries were 
initially grouped into 10 kDalton intact molecular weight 
intervals. For each 10 kDalton protein interval, peptide fragment 
molecular weights were assigned to cells of 100 Dalton intervals. 
The cells therefore contained the number of times a particular 
fragment molecular weight occurred in a protein of any given size. 
This operation was performed for each enzyme. Cell frequency 
values were calculated by dividing each cell value by the total 
number of peptides in each 10 kD protein interval. Cell frequency 
values for each 10 kDalton interval were then normalised to the 
largest cell value (Fmax), with all the cell values recalculated 

	Cell value = Old value / Fmax

to yield floating point numbers between 0 and 1. These 
distribution frequency values, calculated for each cleavage 
reagent, were then built into the MOWSE search program. For 
every database entry scanned, all matching fragments contribute to 
the final score. In the current implementation, non-matching 
fragments are ignored (neutral). For each matching peptide Mw a 
score is assigned by looking up the appropriate normalised 
distribution frequency value. In the case of multiple 'hits' in 
any one target protein (i.e. more than one matching peptide Mw), 
the distribution frequency scores are multiplied. The final 
product score is inverted and then normalised to an 'average' 
protein Mw of 50 kDaltons to reduce the influence of random score 
accumulation in large proteins (>200 kDaltons). The final score is 
thus calculated as:

Score = 50/(Pn x H)

Where Pn is the product of n distribution scores and H the 'hit'
protein molecular weight in kD.

		Important consequences of this type of scoring scheme 
are that matches with peptides of higher Mw carry more scoring 
weight, and that the non-random distribution of fragment molecular 
weights in proteins of different sizes is compensated for.

[6.7] Simulation studies.

	In a simulation of scoring properties, 100 test proteins with 
masses between 10 kD and 100 kD were randomly selected from the 
OWL sequence database. The sets of all possible tryptic peptide 
masses for each protein were randomized and database searches 
performed with increasing numbers of fragments  (default search 
parameters) until the test protein reached the top of the ranked 
scoring list. 99% of the test proteins were correctly identified 
using only five peptides or less (mean=3.6 peptides),  with one 
example requiring six. These figures were surprisingly small 
considering that some of the proteins in the test sample generated 
more than 100 possible tryptic fragments. All 100 test examples were
identified using 30% or less of the maximum number of available peptides.
	This distribution was somewhat dependent on protein size, as 
smaller proteins generally yield fewer peptide fragments. Thus, 
all proteins of 30 kD and over were identified using 13% or less 
of possible fragments (1 in 8), with all proteins of 40 kD and 
above requiring less than 10% (1 in 10). In this latter group, 
therefore, more than 90% of the potential information (all 
possible peptides) was redundant. For the simulation a 'unique' 
identification required matching not only of protein type (e.g. 
globin) but correct discrimination of type, species, and isoform 
or isozyme. Discrimination could be further improved by reducing 
the error tolerance to only +/- 1 Dalton (mean=2.7 peptides). Such 
accuracies are easily bettered using more sophisticated 
ESI/quadrupole or high-field FAB
spectrometers, but the default search value (+/- 2 Daltons) 
compensates for the reduced accuracy obtainable from the smaller 
time-of-flight (TOF) instruments. Mass accuracies better than +/- 1 
Dalton were not essential, and in fact the error tolerance could be 
relaxed to +/- 5 Daltons in many cases with little degradation in 
performance. The simulation thus clearly demonstrated the high 
degree of discrimination afforded by relatively few peptide 
masses, even with generous allowance for error.

[7] General references.

1: Barber M, Bordoli RS, Sedgwick RD, Tyler AN: Fast atom 
bombardment of solids: a new ion source for mass spectrometry. J 
Chem Soc Chem Commun 1981, 7: 325-327.

2: Dole M, Mack LL, Hines RL, Mobley RC, Ferguson LD, Alice MB: 
Molecular beams of macroions. J Chem Phys 1968, 49:2240-2249.

3: Meng CK, Mann M, Fenn JB: Of protons or proteins.  Z Phys D 
1988, 10: 361-368.

4: Karas M, Hillenkamp F: Laser desorption ionisation of proteins 
with molecular masses exceeding 10,000 Daltons. Analytical 
Chemistry 1988, 60:2299-2301.

5: Morris H, Panico M, Taylor GW: FAB-mapping of recombinant-DNA 
protein products. Biochem Biophys Res Commun 1983, 117:299-305.

6: Morris H, Greer FM: Mass spectrometry of natural and 
recombinant proteins and glycoproteins. Trends in Biotechnology 
1988, 6:140-147.

7: Weissenbach J, Gyapay G, Dib C, Vignal J, Morissette J, 
Millasseau P, Vaysseix G, Lathrop M: A second generation linkage 
map of the human genome. Nature 1992, 359:794-801.

8: Adams MD, Kelley JM, Gocayne JD, Dubnick M, Polymeropoulos MH, 
Xiao H, Merril CR, Wu A, Olde B, Moreno RF, Kerlavage AR, McCombie 
WR, Venter JC: Complementary DNA sequencing: expressed sequence 
tags and human genome project. Science 1991, 252:1651-1656.

9: Lehrach H, Drmanac R, Hoheisel J, Larin Z, Lennon G, Monaco AP, 
Nizetic D, Zehetner G, Poustka A: Hybridization fingerprinting in 
genome mapping and sequencing. In Genome Analysis Volume 1: 
Genetic and Physical Mapping. Cold Spring Harbor Laboratory Press; 
1990:39-81 .

10: Akrigg D, Bleasby AJ, Dix NIM, Findlay JBC, North ACT, Parry-
Smith D, Wootton JC, Blundell TI, Gardner SP, Hayes F, Sternberg 
MJE, Thornton JM, Tickle IJ, Murray-Rust P: A protein 
sequence/structure database. Nature 1988, 335:745-746.

11: Bleasby AJ, Wootton JC: Construction of validated, non-
redundant composite protein databases. Protein Engineering 1990, 

More information about the Bionews mailing list