This invention relates generally to analysis of data collected by analytical techniques such as chromatography and spectrometry. More particularly, it relates to a peak detection method for multidimensional data such as liquid chromatography-mass spectrometry (LC-MS) data.
Liquid chromatography-mass spectrometry (LC-MS) is a well-known combined analytical technique for separation and identification of chemical mixtures. Components of a mixture pass through a chromatographic column at different rates, and the eluent is subjected to mass spectrometric analysis at known time intervals. Data are acquired as a series of time-dependent mass spectra, i.e., ion intensity at varying mass-to-charge ratios (m/z).
LC-MS data are typically reported by the mass spectrometer as a total ion current (TIC) chromatogram, the sum of all detected ions at each scan time. A TIC chromatogram of a proteolytic digest of human serum is shown in
LC-MS has been used traditionally to study relatively simple samples, characterized by large available volumes and small numbers of samples and mixture components, leading to spectra containing few peaks. Recently, the method has been applied to proteomic and metabolomic profiling of complex biological mixtures. In such studies, many samples, each containing a large number of components, are analyzed rapidly, and large amounts of data are collected for mining and statistical analysis. While mass spectra acquired in traditional experimental studies can be interpreted manually, high-throughput studies require automated selection of peaks. Because spectra of complex biological samples tend to contain a large number of overlapping peaks and be very noisy, accurate automatic peak picking is a difficult problem to address. State-of-the-art LC-MS instruments provide a basic peak picking function for both chromatograms and mass spectra. A noise threshold level is determined automatically, and local maxima in clusters of points above the threshold are identified as peaks. The operator can instead specify a threshold above which the system designates local maxima as peaks. In other methods, the base peak, i.e., the highest peak in the chromatogram, is identified, and all points whose intensities exceed a preset fraction of the base peak are identified as peaks. In practice, however, although the peak picking is performed automatically, it is not intended to be fully automated, but rather to serve as an aid to the operator in analyzing and interpreting the data. Note also that LC-MS data is two-dimensional; that is, a discrete data point (intensity) is obtained for varying values of two independent variables, retention time and mass-to-charge ratio (m/z). Commercial1y available peak picking methods are applied to one-dimensional data only, i.e., individual mass spectra or chromatograms. For example, Waters's MASSLYNX™ and Thermo Finnigan's XCALIBUR™ are LC-MS software packages that have a peak selection feature.
A method for filtering and recognizing peaks in spectrometry data is disclosed in U.S. Pat. No. 5,995,989, issued to Gedcke et al. An average background signal level and an average deviation from the background are computed and used to define a local threshold value for each point. Points exceeding the threshold are assumed to be peaks or near peaks. This method was developed for mass spectra, and therefore provides a one-dimensional peak recognition algorithm only. Although it can be applied to two—and higher-dimensional data such as two-dimensional LC-MS data, e.g., by selecting peaks in each mass spectrum and then combining the resulting mass spectra into a total ion current chromatogram, the method's analysis remains one-dimensional. Such a method is limited because it does not take advantage of the information provided by the chromatography dimension. That is, what appears to be a peak in a single mass spectrum may be below the noise threshold in a corresponding mass chromatogram (ion abundance versus retention time for a particular m/z value).
More extensive and multidimensional peak picking algorithms have been developed for nuclear magnetic resonance (NMR) spectroscopy, in which manual peak selection is more time consuming and therefore provides a greater incentive for automation. For example, an automated peak picking algorithm for multidimensional NMR spectra is disclosed in R. Koradi et al., “Automated Peak Picking and Peak Integration in Macromolecular NMR Spectra Using AUTOPSY,” J. Magn. Reson., 13: 288–297 (1998). In one feature of the AUTOPSY (automated peak picking for NMR spectroscopy) algorithm, a different local noise level is defined for each multidimensional data point, and the point is retained only if its value exceeds the local noise level. A given point's local noise level is a function of the average noise level for all one-dimensional slices passing through the point.
Existing multidimensional peak picking algorithms, such as AUTOPSY, are generally not sufficiently effective at selecting peaks in LC-MS data to allow fully automated peak detection. Moreover, various aspects of LC-MS data render it not amenable to such analysis. For example, LC-MS noise arises from a variety of unrelated and, in some cases, poorly understood sources, and is therefore difficult to filter effectively using methods developed for noise with well-known distributions, such as those found in NMR data. In addition, it is not uncommon for one mass-to-charge ratio to be very noisy, while a single retention time shows little noise. Using the AUTOPSY algorithm, however, a point at the intersection of the noisy mass and low-noise retention time has a threshold level determined equally by the noise level within each one-dimensional slice. Such a level may be too low to exclude all of the noise. This algorithm and other available methods are therefore not optimal for peak peaking in LC-MS data.
There is still a need, therefore, for an automated peak picking algorithm for LC-MS and other multidimensional data.
The present invention provides a method for automatic peak selection in multidimensional data. Each point of an n-dimensional data set has up to n different one-dimensional peak selection criteria applied to it, and a point can be selected as a peak only if it satisfies all (or some threshold number) of the applied one-dimensional criteria. The method is effective at selecting peaks in very noisy multidimensional data, as well as in data having non-normally distributed noise or noise levels that differ dramatically in different dimensions. As a result, the method facilitates automation of high-throughput analysis of small volumes of biological fluids by two-dimensional analytical techniques such as liquid chromatography-mass spectrometry.
In one embodiment, the invention provides a method for automatically detecting peaks in an n-dimensional data set by independently applying m one-dimensional selection criteria to each point in the data set, where 2≦m≦n. Next, candidate peaks are identified in the points, i.e., those points that satisfy p of the selection criteria applied to them, where 2≦p≦m. Preferably, candidate peaks satisfy m or n applied selection criteria. Each one-dimensional selection criterion applied to a point corresponds to one of the n dimensions of the data set. Preferably, the selection criteria are noise thresholds, and a point whose value exceeds a noise threshold satisfies the criterion. Noise thresholds are computed from a subset of the data points, and preferably from a median of points in the subset. A particular point's criteria are preferably specific to the point, i.e., locally computed from the n (or fewer) one-dimensional components of the data set that contain the point. Thus each dimension has multiple corresponding selection criteria, and different data points may have different criteria applied to them.
After candidate peaks have been selected, a peak recognition algorithm is applied to select actual peaks from the candidate peaks. Preferably, peaks are clusters of contiguous candidate peaks, because peaks generally have an associated width, but additional peak recognition algorithms, such as lineshape analysis or Bayesian/Maximum Likelihood analysis (for mass chromatograms) or isotope distribution analysis (for mass spectra), can also be applied.
The present invention also provides a method for detecting components in a chemical mixture, by subjecting the mixture to chromatography and mass spectrometry using an instrument, acquiring a two-dimensional data set from the instrument, and analyzing the acquired data to identify candidate peaks. The data consist of mass spectra and mass chromatograms, and local noise thresholds are computed for each spectrum and chromatogram. Both corresponding thresholds are applied to each point, and candidate peaks are identified as points exceeding both thresholds. Noise thresholds are preferably computed from a median of the nonzero points in the chromatogram or spectrum.
Also provided by the present invention is a program storage device accessible by a processor, tangibly embodying a program of instructions executable by the processor to perform method steps for an automatic peak detection method as described above.
The present invention provides a method for automatic peak selection in multidimensional data. For illustration purposes, the invention will be described with respect to two-dimensional liquid chromatography-mass spectrometry (LC-MS) data. However, it is to be understood that the invention can be applied to any type of multidimensional data (i.e., two or more dimensions). The invention is particularly well suited to data that has different noise characteristics in different dimensions, or that has non-normally distributed noise, and that therefore cannot be processed effectively using standard multidimensional peak picking algorithms. However, the invention can equally well be applied to data characterized by normally distributed noise in all dimensions.
As used herein, multidimensional data refers to data that can be understood or characterized as having multiple independent variables and a single dependent variable. The dimensionality of the data refers to the number of independent variables. For example, in the case of two-dimensional LC-MS data, the two dimensions are retention time and mass-to-charge ratio (m/z), and an ion intensity level is measured for each mass-to-charge ratio at each retention time. In this case, m/z is not a straightforward independent variable, but it can be treated as such for analysis purposes. Similar decisions can be made to apply the present invention to other data sets that are not straightforward n-dimensional data (n independent variables and one dependent variable). In fact, any data having points with at least three components, for which it is desired to select peaks in one of the components, can be analyzed using the present invention.
Peak picking (also referred to as peak detection, extraction, or selection) refers to identification of actual peaks within multidimensional data that contains noise. In data collected from an analytical instrument, peaks correspond to maxima in physical measurements containing information about the analyte. In general, peaks have a width defined by more than one data point. With respect to the measurement being made, noise has no information content. In general, noise is randomly distributed according to a measurable statistical distribution. Thus in principle, if the noise distribution is known, a suitable filter can be constructed to reduce the noise sufficiently so that peaks can be detected. In general, the peaks may be detected above the noise level provided the average intensity level of the noise is below that of the peaks.
LC-MS noise originates from a variety of sources corresponding to different components of the system. For example, chemical noise results from column bleed, i.e., long-time elution of strongly-adsorbed species at particular mass-to-charge ratios; low-concentration sample contaminants; and detection of the chromatographic mobile phase. In the mass spectrometer, the ion generation, selection, and detection processes all generate noise. Electronic signal processing and analog-to-digital conversion add white noise to the acquired data. The noise sources and distributions are not well understood for all components, making it difficult to threshold the data correctly for peak selection. In fact, as determined by the present inventor, LC-MS noise is not normally distributed, unlike, e.g., purely electrical noise. It therefore cannot be filtered effectively using standard linear filters and peak extraction algorithms. Additionally, the different noise sources can have completely different effects in the two dimensions. For example, the chromatographic mobile phase can corrupt a particular mass chromatogram (abundance versus retention time for one n/z value), making it very noisy, while the effect on the mass spectra (abundance versus m/z at one retention time) is noise only at that one mass-to-charge ratio.
The present invention addresses this nature of LC-MS data by applying different noise thresholds to different dimensions of the data. A flow diagram of a general peak selection method 20 of the invention applied to an n-dimensional data set is shown in
Application of the method to a two-dimensional data set 32 is illustrated in
As shown in
The one-dimensional peak selection criteria are preferably noise thresholds, and points satisfy a criterion by exceeding the noise threshold. n noise thresholds are applied to each data point, and the point is considered to be a candidate peak only if it exceeds all n noise thresholds. The noise thresholds are computed locally, i.e., for each component vector. For example, the noise threshold of the vertical selection criterion 34 in
Preferably, the noise threshold is computed not from a mean value of the data points, but from a median value of the nonzero data points in the data set component. A median is preferred because it is insensitive to outliers and also relatively insensitive to the number of peaks in the subset of which the median is computed. It is therefore not affected by high-intensity peaks and accurately reflects the background noise level. In fact, any method for computing the threshold that is insensitive to outliers can be used. In a simple embodiment, the noise threshold is a multiplicative factor (e.g., between 6 and 16) of the median. Generally, a suitable multiplicative factor can be determined by trial and error and evaluated by the number and quality of resulting candidate peaks. The optimal number typically depends upon the particular application. Of course, the present invention can be implemented with any method for computing the noise threshold. A point whose value exceeds the noise threshold is considered to be a one-dimensional peak, i.e., a peak in that particular dimension.
A flow diagram of a preferred embodiment of a method 40 for implementing the present invention is shown in
Once the vector is selected, in step 48, a noise level is computed for that vector from all points through which the vector passes. In step 50, one-dimensional peaks, i.e., points on the vector whose value exceeds the computed local noise level, are identified. If the point qualifies as a one-dimensional peak, then its value is stored for future processing. Alternatively, a local noise threshold can be computed for each point on the vector from, for example, a window of points surrounding the point, to determine whether the point is a one-dimensional peak. After all data points on the vector have been examined, the system determines whether additional vectors parallel to the selected axis and passing through data points are available (step 52). If so, the method returns to step 46 to select the next vector, e.g., vector 76 of
Once all vectors for a given dimension have been examined, the method determines, in step 54, whether additional unanalyzed dimensions exist. If so, the method returns to step 44 to select vectors parallel to the new axis. As shown in
A peak recognition algorithm is applied to the candidate peaks in step 58 to select actual peaks from the candidate peaks in step 60. For example, in a preferred embodiment, contiguous candidate peaks are clustered, and a minimum number of clustered candidate peaks is considered to be a peak. A variety of clustering algorithms are known in the art; one suitable example is a recursive grassfire algorithm. Alternatively, more sophisticated peak recognition methods such as lineshape analysis or Bayesian/Maximum Likelihood analysis (for mass chromatograms) or isotope distribution analysis (for mass spectra) can be applied before, after, or during clustering. These methods are well-known in the art and will not be discussed herein.
Note that even though the data set may be acquired as (n+1)-dimensional points, it can easily be structured into an n-dimensional matrix to which the present invention is applied. Components lacking values for particular independent variables can simply be considered to have zero values at those independent variables.
In alternative embodiments of the invention, for data sets in which n>2, a point can qualify as a candidate peak even if it does not satisfy all n one-dimensional peak selection criteria. In fact, the invention includes methods in which candidate peaks satisfy two or more one-dimensional selection criteria. The number of dimensions in which the point must be a peak can be predetermined by the user and incorporated into the method. In these embodiments, m selection criteria are applied to each point, where 2≦m≦n, and candidate peaks satisfy p one-dimensional selection criteria, where 2≦p≦m. Note that each candidate peak is not required to satisfy criteria corresponding to the same dimension. For example, in a seven-dimensional data set, each candidate point can be required to satisfy conditions corresponding to the first five dimensions, in which case only these conditions need be applied to each point. Alternatively, a candidate peak can be required to satisfy any five of seven applied criteria. Any combination of values of m and p as defined above is within the scope of the invention. An application of the alternative embodiments is the combined use of three analytical techniques, e.g., LC, MS, and NMR spectroscopy. In that case, candidate peaks may be tested and qualify as one-dimensional peaks in the LC and MS dimensions only.
As with most peak selection methods, higher resolution is beneficial for clustering or post-clustering filtering of identified peaks. Higher resolution scans provide more data points for each true peak, making the peak easier to identify. With high enough resolution in the mass spectra, additional criteria can be applied. For example, isotope analysis examines m, m+1, and m+2 peaks to compare their distribution to a known isotope distribution. If the peaks have the expected distribution, then they can be selected with more confidence.
The above methods can be modified in many ways without departing from the scope of the invention. For example, a baseline correction can be applied to the data before the peak picking method is applied. The baseline correction can, for example, compensate for the low-frequency upward drift in the chromatogram over time as the mobile phase concentration changes, thereby affecting the ionization efficiency in the mass spectrometer. The data can also be smoothed, filtered, or centroided before the method is applied. Centroided data replaces the entire lineshape of a peak with a point at the average value of the independent variables in multidimensional space. In fact, data obtained from a mass spectrometer is often not raw data, but has already been centroided or otherwise modified before made available to the user.
The methods of the invention can also be used as one step of a peak selection process. For example, after peaks have been selected from the candidate peaks using the present invention, additional criteria can be applied to the peaks before they are accepted and stored in a peak database. Alternatively, the method of the invention can be applied after an initial peak finding algorithm has been applied. In this case, the method of the invention confirms that the selected peaks satisfy a peak selection criterion in all dimensions.
Note that although the method has been described for LC-MS data, it can be applied to any multidimensional data in which peaks are detected. In particular, data collected by other chromatographies (e.g., gas), spectroscopies, or capillary electrophoresis would benefit from methods of the invention. For example, LC-NMR is a hyphenated method combining chromatography with nuclear magnetic resonance spectroscopy. LC-LC-MS methods add an additional chromatography dimension and are well suited for analysis by the present invention.
Although not limited to any particular hardware configuration, the present invention is typically implemented in software by a system 90, shown in
The computer 92 implementing the invention typically contains a processor 102, memory 104, data storage medium 106, display 108, and input device 110. Methods of the invention are executed by the processor 102 under the direction of computer program code stored in the computer 92. Using techniques well known in the computer arts, such code is tangibly embodied within a computer program storage device accessible by the processor 102, e.g., within system memory 104 or on a computer readable storage medium 106 such as a hard disk or CD-ROM. The methods may be implemented by any means known in the art. For example, any number of computer programming languages, such as Java, C++, or LISP may be used. Furthermore, various programming approaches such as procedural or object oriented may be employed.
It is to be understood that the steps described above are highly simplified versions of the actual processing performed by the computer 92, and that methods containing additional steps or rearrangement of the steps described are within the scope of the present invention.
It should be noted that the foregoing description is only illustrative of the invention. Various alternatives and modifications can be devised by those skilled in the art without departing from the invention. Accordingly, the present invention is intended to embrace all such alternatives, modifications and variances which fall within the scope of the disclosed invention.
This application is a continuation application of U.S. patent application Ser. No. 10/226,090, filed Aug. 22,2002, now U.S. Pat. No. 6,873,915, which claimed the benefit of U.S. Provisional Application No. 60/314,841, filed Aug. 24, 2001. Both applications are incorporated herein by reference.
Number | Name | Date | Kind |
---|---|---|---|
3552865 | Leung et al. | Jan 1971 | A |
3690836 | Buissiere et al. | Sep 1972 | A |
3997298 | McLafferty et al. | Dec 1976 | A |
3999047 | Green | Dec 1976 | A |
4405235 | Rossiter | Sep 1983 | A |
4426451 | Columbus | Jan 1984 | A |
4643570 | Machler et al. | Feb 1987 | A |
4752888 | Yoshihara | Jun 1988 | A |
4761381 | Blatt et al. | Aug 1988 | A |
4786813 | Svanberg et al. | Nov 1988 | A |
4844617 | Kelderman et al. | Jul 1989 | A |
4963498 | Hillman et al. | Oct 1990 | A |
5072382 | Kamentsky | Dec 1991 | A |
5091652 | Mathies et al. | Feb 1992 | A |
5119315 | Kemp et al. | Jun 1992 | A |
5127730 | Brelje et al. | Jul 1992 | A |
5192980 | Dixon et al. | Mar 1993 | A |
5239178 | Derndinger et al. | Aug 1993 | A |
5274240 | Mathies et al. | Dec 1993 | A |
5304810 | Amos | Apr 1994 | A |
5377003 | Lewis et al. | Dec 1994 | A |
5412208 | Covey et al. | May 1995 | A |
5430542 | Shepherd | Jul 1995 | A |
5446532 | Yamazaki | Aug 1995 | A |
5453505 | Lee et al. | Sep 1995 | A |
5456252 | Vari et al. | Oct 1995 | A |
D366938 | Shartle et al. | Feb 1996 | S |
5492833 | Rodriguez et al. | Feb 1996 | A |
5523573 | Hanninen et al. | Jun 1996 | A |
5532873 | Dixon | Jul 1996 | A |
5547849 | Baer et al. | Aug 1996 | A |
5556764 | Sizto et al. | Sep 1996 | A |
5578832 | Trulson et al. | Nov 1996 | A |
5585246 | Dubrow et al. | Dec 1996 | A |
5592402 | Beebe et al. | Jan 1997 | A |
5627041 | Shartle | May 1997 | A |
D382648 | Shartle et al. | Aug 1997 | S |
5658735 | Lee | Aug 1997 | A |
D383852 | Shartle et al. | Sep 1997 | S |
5672869 | Windig et al. | Sep 1997 | A |
5682038 | Hoffman | Oct 1997 | A |
5687964 | Stephan et al. | Nov 1997 | A |
5689110 | Dietz et al. | Nov 1997 | A |
5692220 | Diamond et al. | Nov 1997 | A |
5710713 | Wright et al. | Jan 1998 | A |
D391373 | Shartle | Feb 1998 | S |
5713364 | DeBaryshe et al. | Feb 1998 | A |
5726751 | Altendorf et al. | Mar 1998 | A |
5734058 | Lee | Mar 1998 | A |
5736410 | Zarling et al. | Apr 1998 | A |
5739000 | Bierre et al. | Apr 1998 | A |
5741411 | Yeung et al. | Apr 1998 | A |
D395708 | Shartle | Jun 1998 | S |
5795729 | Lee | Aug 1998 | A |
5814820 | Dong et al. | Sep 1998 | A |
5832826 | Mack et al. | Nov 1998 | A |
5867610 | Lee et al. | Feb 1999 | A |
5871946 | Lucas et al. | Feb 1999 | A |
5885841 | Higgs, Jr. et al. | Mar 1999 | A |
5910287 | Cassin et al. | Jun 1999 | A |
5912134 | Shartle | Jun 1999 | A |
5932428 | Dubrow et al. | Aug 1999 | A |
5962238 | Sizto et al. | Oct 1999 | A |
5981180 | Chandler et al. | Nov 1999 | A |
5995989 | Gedcke et al. | Nov 1999 | A |
6002986 | Mito | Dec 1999 | A |
6008490 | Kato | Dec 1999 | A |
6008896 | Sabsabi et al. | Dec 1999 | A |
6017693 | Yates et al. | Jan 2000 | A |
6059724 | Campbell | May 2000 | A |
6063338 | Pham et al. | May 2000 | A |
6066216 | Ruppel, Jr. | May 2000 | A |
6072624 | Dixon et al. | Jun 2000 | A |
6091492 | Strickland et al. | Jul 2000 | A |
6093573 | Beamer et al. | Jul 2000 | A |
6104945 | Modell et al. | Aug 2000 | A |
6112161 | Dryden et al. | Aug 2000 | A |
6133046 | Clerc | Oct 2000 | A |
6134002 | Stimson et al. | Oct 2000 | A |
6138117 | Bayardo | Oct 2000 | A |
6147344 | Annis et al. | Nov 2000 | A |
6200532 | Wu et al. | Mar 2001 | B1 |
6207955 | Wells et al. | Mar 2001 | B1 |
6215892 | Douglass et al. | Apr 2001 | B1 |
6229603 | Coassin et al. | May 2001 | B1 |
6229635 | Wulf | May 2001 | B1 |
6232114 | Coassin et al. | May 2001 | B1 |
6236945 | Simpson et al. | May 2001 | B1 |
6253162 | Jarman et al. | Jun 2001 | B1 |
6278794 | Parekh et al. | Aug 2001 | B1 |
6334099 | Grace et al. | Dec 2001 | B1 |
6376843 | Palo | Apr 2002 | B1 |
6377842 | Pogue et al. | Apr 2002 | B1 |
6388788 | Harris et al. | May 2002 | B1 |
6391649 | Chait et al. | May 2002 | B1 |
6400487 | Harris et al. | Jun 2002 | B1 |
6421612 | Agrafiotis et al. | Jul 2002 | B1 |
6449584 | Bertrand et al. | Sep 2002 | B1 |
6514767 | Natan | Feb 2003 | B1 |
6526299 | Pickard | Feb 2003 | B2 |
6552784 | Dietz et al. | Apr 2003 | B1 |
6590204 | Baranov | Jul 2003 | B2 |
6603537 | Dietz et al. | Aug 2003 | B1 |
6620591 | Dunlay et al. | Sep 2003 | B1 |
6625546 | Sepetov et al. | Sep 2003 | B2 |
6642059 | Chait et al. | Nov 2003 | B2 |
6646271 | Yokokawa et al. | Nov 2003 | B2 |
6687395 | Dietz et al. | Feb 2004 | B1 |
6753966 | Von Rosenberg | Jun 2004 | B2 |
6787761 | Hastings | Sep 2004 | B2 |
6800860 | Dietz et al. | Oct 2004 | B2 |
6835927 | Becker et al. | Dec 2004 | B2 |
6858435 | Chervet et al. | Feb 2005 | B2 |
6873915 | Hastings | Mar 2005 | B2 |
6937330 | Dietz et al. | Aug 2005 | B2 |
6950185 | Da Silva et al. | Sep 2005 | B1 |
6962818 | Schneider et al. | Nov 2005 | B2 |
6979830 | Dietz et al. | Dec 2005 | B2 |
20010019829 | Nelson et al. | Sep 2001 | A1 |
20020049152 | Nock et al. | Apr 2002 | A1 |
20020053545 | Greef | May 2002 | A1 |
20020095419 | Parce | Jul 2002 | A1 |
20020102610 | Townsend et al. | Aug 2002 | A1 |
20020123055 | Estell et al. | Sep 2002 | A1 |
20020141051 | Vogt et al. | Oct 2002 | A1 |
20030087322 | Aebersold et al. | May 2003 | A9 |
20040257576 | Kirsch et al. | Dec 2004 | A1 |
20060000984 | Wolleschensky et al. | Jan 2006 | A1 |
Number | Date | Country |
---|---|---|
0296136 | Dec 1988 | EP |
0421736 | Apr 1991 | EP |
0681177 | Nov 1995 | EP |
0969283 | Jan 2000 | EP |
1407247 | Sep 1975 | GB |
WO-97 43732 | Nov 1997 | WO |
WO 981661 | Apr 1998 | WO |
WO 0067017 | Nov 2000 | WO |
WO-00 70340 | Nov 2000 | WO |
WO 0135266 | May 2001 | WO |
WO-01 44269 | Jun 2001 | WO |
WO-2088819 | Nov 2002 | WO |
Number | Date | Country | |
---|---|---|---|
20050209789 A1 | Sep 2005 | US |
Number | Date | Country | |
---|---|---|---|
60314841 | Aug 2001 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 10226090 | Aug 2002 | US |
Child | 11075370 | US |