WO1997001751A1

WO1997001751A1 - Determination of component concentrations taking account of measurement errors

Info

Publication number: WO1997001751A1
Application number: PCT/IB1996/000616
Authority: WO
Inventors: Ronald S. Scharlack; Lester Sodickson
Original assignee: Chiron Diagnostics Corporation
Priority date: 1995-06-29
Filing date: 1996-06-27
Publication date: 1997-01-16
Also published as: DE69604923T2; EP0835438B1; EP0835438A1; CA2224994A1; DE69604923D1; KR19990028482A; ATE186117T1; US5724268A; BR9608712A; AU705578B2; PL324314A1; AU6013296A

Abstract

The present invention provides apparatus and methods for determining the concentration of sample components of a sample by an analytical technique that yields a spectrum that can be written as Y(φ) = P(φ) . C. The apparatus and methods of the invention account for experimental errors that give rise to distortions in the observed spectrum and that consequently result in inaccurate determinations of sample component concentrations. The invention accounts for such errors by modeling the total experimental error as the sum of one or more types of errors that can be written as κ . K. The spectrum is then modeled as Y = P . C + κ . K. Using the observed spectrum, known values for P, and a mathematical model for κ, this equation can be solved for the best fit value of the sample component concentrations, C, and the magnitudes of the errors, K. The method can be used for any error that can be modeled in the foregoing manner, such as a shift in the spectrum. Particular types of shift include constant shift as well as linear shift across the entire spectrum. The apparatus and methods are advantageously used in absorbance spectroscopy and chromatography.

Description

DETERMINATION OF COMPONENT CONCENTRATIONS TAKING ACCOUNT OF MEASUREMENT ERRORS.

BACKGROUND

Field of the Invention

The present invention relates to the field of analytical spectrometry and, in particular, to apparatus and methods of correcting for deteπnining analytical sample component concentrations that account for instrumental error. Summary of the Related Art

A wide variety of analytical techniques have been developed over the years to detect and determine the concentrations of components of a sample. Some techniques are entirely spectrophotometric, such as ultraviolet-visible-infrared (UV-VIS-IR) absorbance spectroscopy and NMR spectroscopy. Other techniques, such as column chromatography, are not spectrophotometric per se, but often use spectrophotometric techniques to detect the presence of compounds.

Each of these techniques generally provides a "spectrum'' in which a dependent variable, typically the intensity of some quantity (e.g., absorbance), is plotted against a dependent variable (e.g., wavelength). Relative concentrations of sample components are determined by obtaining the best fit to the experimental spectrum by varying the relative spectral contribution of each component. This requires knowledge of the spectral features of the individual components. Owing to instrumental errors and/or deviations in experimental conditions, small shifts of the independent variable often occur that can lead to large changes in the relative concentrations of sample components determined from the measured spectra. To date, the prior art is devoid of suitable methods for determining the shift and correcting for it. Accordingly, new methods of accounting for shifts of the independent variable in analytical spectrophotometric techniques are desirable. SUMMARY OF THE INVENTION

The present invention provides apparatus and methods for the accurate determination of sample component concentrations. The apparatus and methods of the invention advantageously correct for experimental errors (induding instrument induced errors) that would otherwise introduce errors into the measured sample component concentrations. The present invention compensates for a wide variety of phenomena exogenous to the analytical sample that contribute to and are manifested in the observed analytical spectrum, from which sample component concentrations are determmed.

The present invention can be employed advantageously in analytical spectrophotometric techmques such as UV-VIS-IR spectroscopy and NMR spectroscopy. The present invention can also be used in analytical techniques that are not spectrophotometric per se, but which incorporate spectrophotometric detection and/or yield a "spectrum-like" graph (i.e. , a graph that resembles a spectrum). An example of a technique that yields a "spectrum-like" graph is column chromatography, in which materials are detected by absorbance spectroscopy at one (or more) wavelength, yielding a "spectrum-like" graph that depicts the intensity of the eluate absorbance as a function of time. As used herein, the term "spectrum" encompasses traditional spectrophotometric spectrum wherein a spectral intensity is plotted against radiation frequency, wavelength, or wavenumber (or some equivalent thereof), as well as "spectrum-like" graphs produced in techniques such as column chromatography.

Under ideal conditions, an analytical spectrum can be described by the equation

Y(ω) = P(ω) ^• C , (i) were ω is a functional parameter (e.g. , frequency, wavelength, wavenumber, or time), Y is a vector whose elements are the spectral intensities, C is a vector whose elements are the concentrations of the sample components, and P is a matrix whose elements are a measure of the magnitude of the contribution of each sample component to the spectral intensity at each value of ω . The elements of P are known quantities and can be, for example, the UN-VIS-IR extinction coefficients of each sample component at each wavelength λ. In practice, when one desires to use such analytical spectra to determine the concentrations of sample components that give rise to the observed spectrum, Y^, equation (i) can be used to obtain a best fit estimate of C by using Y_ob5 in place of Y. Owing to instrumental and other experimental errors, however, equation (i) frequently does not describe the observed spectrum well. In that case, equation (i) can be modified to incorporate a term representing the experimental error:

Y = P ^• C + dY , (ii) where dY is the experimental error-induced deviation of the observed spectrum from the ideal. In the present invention, dY is written as: dY = ξ-K , (iii) where K is a scalar representing the magnitude of the error and ξ is a vector whose elements are the relative errors at each value of ω . Equation (ii) is then written as Y = P C + ξ K . (iv)

The apparatus and methods of the present invention can be used to correct for experimental errors whenever the observed spectrum can be written in the form of equation (iv), i.e., whenever the experimental error can be written as in equation (iii). Equation (iv) is readily solved for the best fit sample component concentrations, C, and the magnitude of the error, K, to the observed spectrum, Y_ob,, (using Y^, in place of Y in equation (iv)) as described more fully below.

In one aspect of the invention, the apparatus and methods model the error, dY, as a shift of the entire spectrum by an amount dω . In this aspect of the invention: θY dY = — dω = Y ' dω . (v)

8ω In this aspect of the invention, the observed spectrum is estimated as:

Y*. = P C + Y'dω . (vi)

Depending on the model chosen for the shift, dω can be a scalar (i.e. , the same for all ω) or a vector whose elements vary with ω .

In one embodiment of this aspect of the invention, using an appropriate model for dω, equation (vi) takes the form of equation (iv), which is then solved for the best fit values of C and K.

In another embodiment of this aspect of the invention, the apparatus and methods account for shift due to experimental error by adjusting the entire spectrum (to yield an adjusted spectrum, Y_adj) using some weighted average dω of previously determined values of the shift, dω. The entire spectrum is then adjusted by dω :

Y_adj = Y_ob (ω + dω ) . (vii) When dω is small, Y' dω is a good estimate of the shift in the spectrum, which is then preferably calculated from:

On the other hand, when dω is large, it is preferable to account for the shift by using equation (vii) directly. In either case, in this embodiment of the mvention, Y^ is used in either equation (i) or (vi) to obtain the best fit value of C.

In another embodiment of this aspect of the invention, the form of Y in equation (i) is used in equation (v) to obtain an expression for dY in terms of the derivative to P with respect to ω : dY = ay d.ω = ap-c d .ω = ap -C _d .ω , (ix) dω dω dω and the estimated spectrum is:

Y„_t = P ^• C + P' ^• C dω . (x)

Equation (x) reduces to the form of equation (iv) when dω is modeled appropriately and can then be solved for C. In another embodiment of this aspect of the invention, the apparatus and methods of the present invention account for shift adjusting P using some weighted average of previously determined values of dω. An adjusted value of P that accounts for the shift is then given by:

P^ = P (ω + dω ) . (xi) If dω is large, equation (xi) is preferably used directly. If dω is small, however,

(dP/dω)dω is a good estimate of the shift-induced change in P, and P^ is then preferably obtained from:

In either case, P^ is then used in either equation (i) or (x) to obtain the best fit value of C.

In this aspect of the invention, the shift can be modeled in a number of ways. In one embodiment, the shift is modeled as being constant across the entire spectrum. In this model dω = S, a scalar. In another embodiment, the shift is modeled as varying linearly about a central value of ω, ω_c, and is given by the vector whose elements are dω_ι=(ω_ι-ω_c)M , where M is the magnitude of the shift. In this embodiment, dω is a vector. When M> 0 the entire spectrum is magnified about ω_c; when M <0, the entire spectrum is compressed about ω_c. In yet another embodiment, the experimental shift is modeled as a linear combination of the two foregoing models.

As described more fully below, each of these models can account for errors in observed spectrum that are manifested by shifts in ω . The apparatus and methods of the present invention incorporate the foregoing equations to compensate for experimental shift in observed spectra and thereby yield more accurate measurements of the sample component concentrations.

In a particularly preferred embodiment of the present invention, the apparatus and methods of the invention employ UN-NIS-IR absorbance spectroscopy. Co-oximeters, which measure the relative concentrations of blood components, are an example of one embodiment. In this embodiment, Y is the absorbance spectrum A, P is the matrix of extinction coefficients, E, and the independent variable, ω, is the wavelength, λ.

In another preferred embodiment, the apparatus and methods of the invention employ chromatographic means. In this embodiment, ω is me time at which the sample components elute, Y is the absorbance at the wavelength at which the components are detected, and P is a matrix of the relative absorbances of each sample component as a function of elution time.

DETAILED DESCRD7TIOΝ OF THE PREFERRED EMBODIMENTS

The apparatus and methods of the present invention are useful for correcting for a wide variety of instrumental and odier experimental errors in a wide variety of spectrophotometric and analytical techniques. In general, the apparatus and methods of the present invention can be used in conjunction with any spectrophotometric and/or analytical technique yielding a spectrum that can be described by d e equation:

Y(ω) = P(ω) ^• C , (1) where Y is a vector whose "m" components Y_j are the intensities of the spectrum at each of the "m" values of the independent variable ω (G>i), C is a vector whose "n" elements are the concentrations of the sample components d at contribute to the measured response Y, and P is an "m x n" matrix whose elements P_y relate die contribution of component _i to the intensity Y_j. "m" is an integer and equal to me number of values at which Y is measured, "n" is also an integer and equal to the number of sample components contributing to Y. In UV-VIS-IR spectroscopy, for example, Y_χ . is the sample absorbance at wavelength ω_is and die Py are die extinction coefficients of absorbant "j" at wavelength ωj. In column chromatography, as another example, Y-, is the intensity of the absorbance at time ω„ and die P_y are the extinction coefficients of the absorbant "j" at die monitoring wavelength at elution time ω_t under die particular elution conditions (e.g. , buffer identity and strength, pH, temperature, etc.). In practice, when one desires to use such an analytical spectrum to determine ie best estimate of sample components concentrations, equation (1) can be used by inserting Y_ob, for Y and solving for the best fit value of C.

As will become clearer below, it will be appreciated by diose skilled in die art that the present apparatus and memods are ideally suited to analytical techniques in which the identity of die "n" sample components and die matrix elements of P are known and it is desired to determine die best estimate concentrations of the sample components from the observed (measured) spectrum Y_ob,.

The present apparatus and memods can be employed advantageously in analytical spectrophotometric techniques such as UN-VIS-IR spectroscopy and ΝMR spectroscopy. The present methods can also be used in analytical techniques that are not spectrophotometric per se, but which incorporate spectrophotometric detection and/or yield a "spectrum-like" graph (i.e. , a graph that resembles a spectrum). An example of a technique that yields a "spectrum-like" graph is column chromatography, in which materials are detected by absorbance spectroscopy at one (or more) wavelengm, yielding a "spectrum-like" graph that depicts die intensity of die eluate absorbance as a function of time. As used herein, d e term "spectrum" encompasses traditional spectrophotometric spectrum wherein a spectral mtensity is plotted against radiation frequency, wavelengdi, or wavenumber (or some equivalent thereof), as well as "spectrum-like" graphs produced in techniques such as column chromatography. It is frequently the case that errors in die observed spectrum are introduced due to less dian ideal instrument performance and/or experimental technique. As used herein, die term "experimental error" means any error (e.g. , arising from less than ideal instrument performance or suboptimal experimental technique) resulting in a deviation of d e measured spectrum from d e theoretical ideal. These errors translate into errors in the measured sample component concentrations. These errors in the observed spectrum can be accounted for by rewriting equation (1) as Y_ob, = P ^• C + dY , (2) where dY is the experimental error-induced deviation in Y^, from die ideal. The present apparatus and me iods corrects for such errors by incorporating a mathematical model for dY: dY = ξ ^• K (3) where K is a vector of dimension "r" whose elements are the magnitudes of die errors for each of the "r" types of error modeled as contributing to die spectrum, and ξ is an "m x r" matrix comprised of "r" vectors each of whose "m" elements are the relative errors at each value of ω for each type of error modeled as contributing to die spectrum, "r" is an integer and equal to die number of types of errors modeled as contributing to die spectrum. Where only one type of enor is modeled r= l, and K is a scalar and ξ is a vector. Equation (3) is dien written as:

Y_ob, = P ^• C + ξ ^• K (4)

In its broadest aspect, the apparatus and memods of die present invention incorporate corrections for instrumental and/or odier experimental induced errors in measured spectra whenever the spectra can be written in d e form of equation (4).

In diis aspect, therefore, d e invention comprises an improved apparatus and methods of determining the concentrations of sample components using an analytical technique mat yields a spectrum mat can be estimated as equation (1), the improvement comprising correcting for experimental error by modeling die experimental error as "r" types of errors given by the product ξ ^• K, where K is a vector whose "r" elements are the magnitudes of each of the "r" types of experimental errors and ξ is an "m x r" matrix whose elements are the relative errors at each value of ω for each type of experimental error, adding d e product ξ ^• K to me estimated spectrum as in equation (4), and solving for me best fit values of C and K.

As used in die present invention r ≥ 1; preferably r ≤ 10; and most preferably 1 ≤ r ≤ 3. n ≥ 1 and preferably 1 ≤ n ≤ 20. m≥ n + r and preferably m is about twice n + r.

Equation (4) is readily solved for the best fit values of C and K by least squares analysis. To do so, equation (4) is "collapsed" by defining an "m x (n+r)" augmented matrix, P_ξ, which has the form:

and an augmented vector, C_κ, of dimension "n+r," which has the form:

Equation (4) men becomes:

The least squares solution to equation (7) is given by:

where

P_*t = (P . _Pξ)-> . p τ ₍₉₎ is the least squares transfoπnation matrix. E.g. , Noble and Daniel, Applied Linear Algebra, pp. 57-65 (Prentice-Hall, Inc., N.J., 1977). C_κ is readily determined from equation (8) using standard algorid ms. See, e.g., Press et al., Numerical Recipes.- The Art of Scientific Computing (Cambridge University Press, Cambridge 1986). The first "n" elements of C_κ are the best fit values of die "n" sample component concentrations and ie remaining "r" elements are die best fit magnitudes of me errors.

In anodier aspect of the invenuon, the apparatus and memods model die enor, dY, as a shift of ie entire spectrum by an amount dω . In diis aspect of the invention: dY = — dω = Y ' dω . (io) dω In one embodiment of this aspect of me invention, the spectrum is estimated by substituting equation (10) into equation (2) to obtain

Y_obJ = P ^• C + Y'dω . (11) The term Y'dω must be capable of being written in the form of equation (3) and, dierefore, equation (11) in die form of equation (4). As will be demonstrated below, this can be so even when dω itself is other d an a scalar.. In diat instance, dω is decomposed into a scalar K and a vector or matrix and die vector or matrix combined wid Y' to form a product matrix ξ . Equation (11) is tiien solved in die same manner described above for equation (4). Specific examples of this mediod are presented in greater detail below.

In anotiier embodiment of this aspect of the invention, die shift due to experimental error is accounted for by adjusting die entire spectrum using some weighted average of the magnitude of die shift, dω , obtained, for example, from previously determined values of e shift, dω . The entire spectrum is conected for experimental enor by shifting it by an amount dω :

When dω is large relative to die resolution of die measured spectrum, equation (12) is preferably used direcdy to obtain the adjusted, conected spectrum, Y_adJ. When dω is small relative to d e measured spectrum, Y' dω is a good estimate of the shift in the spectrum, which is then preferably calculated from:

Y_adJ = Y^ + Y' ^ . (13)

In either case, however, Y_adj is then used in eitiier equation (1), (11), or (19) (see infra) in place of Y_ob, and die equations solved to obtain the best fit value for C diat is conected for shift. dω is any suitable scalar representing the shift in ω . In a preferred embodiment, dω is the previously calculated value of dω , or an average over die last "k" measurements, where "k" is 2 or more. In a prefened embodiment, K is 5 or more. In anotiier prefened embodiment, k = 8. This method weights each of die last "k" measurements equally. Alternatively, a filter can be used to give greater weight to die most recent values of dω. In d is embodiment, each of the last "k" values of dω are weighted by a factor "w,. " dω is tiien represented by a vector, dω", where each of die "k" elements dω," is a previously determined value of dω, such d at dω,^* is d e most recendy determined value of dω and dω_k ^* is the oldest value, dω is tiien obtained from the equation: dα> = w^{τ •} dω^* , (14) wherein w^τ is the transpose of the vector w whose "k" values "w," are chosen such diat w, ≥ w₂ ≥ . . .≥ w_k and:

Σ i ^{= l} . (15) ι = l w can be determined in any suitable manner. For equal weighting of the last "k" values of dω , each Wj is 1/k. Alternatively, where it is desired to give greater weight to die most recently determined values of dω, a function such as:

can be used, wherein "a" is a real number greater than 0. Preferably "a" is greater dian 1. Odier suitable weighting functions are well known to those skilled in d e art.

In another embodiment of diis aspect of the invention, the form of Y in equation (1) is used in equation (10) to obtain an expression for dY in terms of the derivative to P with respect to ω : dY . dP C . dP _ , _ , _ , dY = dω = dω = -C dω = P C dω n s) dω dω dω ' and the best fit value of C conected for shift is obtained from:

Y_o-,, = P ^• C + P' ^• C dω . (19) Equation (19) reduces to the form of equation (4) when dω is modeled appropriately and an estimated value for C, C^, is used in the expression P' ^• C. C_tιf can be obtained, for example, from the solution to equation (1):

C„. = F- ^• Y_ob* (20) where P^f = (P^{τ •} PV^{1 •} P^τ (21) is the least squares transformation matrix and Y_ob, is die observed (or measured) spectrum.

Widi diis estimate equation (19) is men solved for the best fit value of C and die magnitude of me shift in die same way equation (4) is solved, as described above.

In anotiier embodiment of diis aspect of the invention, the shift is accounted for by adjusting P using some weighted average of previously determined values of dω. An adjusted value of P diat accounts for the shift is tiien given by: P_adj = P (ω + dω ) . (22)

If dω is large, equation (21) is preferably used directiy. If dω is small, however, P' dω is a good estimate of the shift-induced change in P, and P_adj is tiien preferably obtained from: P^ = P + P' dω^" . (23)

In either case, P^ is then used in eitiier equation (1) or (19) in place of P to obtain a value for C that is conected for shift. In either case, P^ is used in any of equations (1), (11), and (19) to determine die magnitude of die shift, dω, and a conected value of C.

The shift can be modeled in a variety of ways. For example, in one embodiment, the shift in ω is modeled as being constant across the entire spectrum. In diis embodiment dω is defined as a scalar S, which can be positive or negative, dω is given by some weighted average, S , of S. In d is embodiment equation (11) takes the form:

Y_ob, = P ^• C + Y' ^• S . (24)

In this embodiment, the shift can be accounted for to obtain more accurate values of C by solving equation (24) in d e manner described for solving equation (4), supra, where ξ =

Y' and K=S. Alternatively, the entire spectrum may be shifted by amount S, preferably as in equation (12) when S is large:

or as in equation (13) when S is small: Y_adj = Y_ob, + Y'S. (26)

In either case, however, Y_adj is then used in eitiier equation (1), (24), or (27) (see infra) in place of Y_ob, to obtain S and a best fit value of C diat is conected for die shift.

Alternatively, the shift is accounted for using the derivative of the matrix P, as in equation (19): Y_ob, = P ^• C + P' ^• C S . (27)

Equation (27) is solved in the same manner as equation (19) to obtain the magnitude of die shift S and a value of C diat is conected for the shift. Or, if a reasonable estimate of S is available, equation (22) can be used to conect for die shift by adjusting the matrix P, using: P_adJ = P (ω + S) . (28) if S is large, or: P^_j = P + P' S . (29) if S is small. In either case, P^_j is tiien used in any of equations (1), (24), and (27) in place of P to obtain a value for C that is conected for shift. If either equation (24) or (27) is used, a new value for the magnitude of die shift, S, is also obtamed. In anotiier embodiment for modeling experimental enor arising from a shift in die spectrum, the shift is modeled as a compression/magnification of the spectrum about a central value of ω, ω_c. The change in ω due to magnification or compression in this model is given by dω, = (co, - ω_c) M , (30) where ω_t is the "i"¹" component of the vector ω, whose elements are the values of ω at which measurements are taken, and M is die magnification compression factor. It is seen that in this model dω is a vector. If M > 0, then the scale of the independent variable ω is magnified. If M <0, tiien die scale is compressed. It is seen from equation (30) diat die change in ω for which this method compensates is directiy proportional not only to the magnification/compression factor M, but also to d e distance from the central value of ω, ω_c. Thus, the greater the distance from the central value, ω_c, the greater the change.

In this embodiment equation (11) takes the form:

Y_ob, = P - C + Y' - Δ M . (31) where Δ is a diagonal matrix whose diagonal elements Δ„ are (ω, - ω_c). Equation (31) is the same as equation (4), with ξ = Y' -Δ and K=M, and can be solved in the same manner.

Alternatively, if a reasonable value M is available (as described in equations (14)- (17) and associated text), the spectrum may be shifted by amount Δ M , preferably as in equation (12) when M is large: Y_adJ., = Y_obs (ω, - (ω, - ω_c)M ) , (32) or as in equation (13) when M is small:

Y_ad„ = Y_ob, + Y' ^• Δ M . (33)

In either case, however, Y_adj is then used in eitiier equation (1), (31), or (34) (see infra) in place of Y _obg to obtain M and a best fit value of C tiiat is conected for the magnification/compression type shift. Alternatively, die magnification/compression type shift is accounted for using d e derivative of die matrix P, as in equation (19):

Y_ob, = P C +Δ P' C M . (34)

Equation (34) is solved in die same manner as equation (19) (by estimating C in the expression Δ ^• P' ^• C, which is equivalent to ξ in equation (4)) to obtain M and a value of C tiiat is conected for the magnification/compression type shift. Or, if a reasonable estimate of M is available (as described in equations (14)-(17) and associated text), equation (34) can be used to conect for the magnification compression type shift by adjusting die matrix P, using: P_ΛΛiλ = -p (ω_l + (ω., - ω_c) M ) . (35) if M is large, or:

P_adj -= P +Δ P' C (36) if M is small. In either case, P^_j is then used in any of equations (1), (31), or (34) in place of P to obtain a best fit value for C that is conected for the magnification/compression type shift. If either equation (31) or (34) is used, a new value of M is also obtained

In yet anotiier aspect of the invention, the apparatus and methods incorporate the foregoing techniques to compensate both for shift and magnification/compression. In this aspect of the invention, d e shift, dω , is given by db>; = S +(ω, - ω_c)M . (37)

Using this expression for dω in equation (11) yields:

Y_ob, = P • C + Y'S 4- Δ ^• Y' M . (38)

Equation (36) can be solved in die very same manner as equation (4) by defining ξ as d e matrix [Y\ Δ ^• Y'] and K as a vector [S,M]. Whereas die previous two embodiments illustrated models to equation (4) in which ξ and K were a vector and a scalar, respectively, this embodiment illustrates a situation in which ξ and K are a matrix and a vector, respectively.

In an altemative embodiment of diis aspect of die invention, a weighted average of die scalars S and M are used to precalculate an adjusted spectrum, Y^, using equation (12) wherein dω is S + (ω, - ω_c)M :

Y_adj = Y_ob*(ω, + S + (ω, - ω_c) M) (39) which is used if die sum S -+- (G>_J - ω_c) M is large, or:

Y_adj = Y + Y' S + Δ ^• Y'M (40) when the sum S + (ω, - ω^ M is small. In either case, Y^ is used in any of equations (1), (38) or (41) in place of Y_ob, to obtain new values for S and M and a best fit value for C that is conected for both types of shift enor.

In another embodiment of diis aspect of the invention, the change in ω due to the combined effects of shift and magnification/compression can be accounted for by calculating die derivative of die matrix P. Using die results previously reported, equations (27) and (34), the estimated spectrum is: Y_ob$ = P C +P - C S +Δ P' C M . (41)

Equation (41) is analogous to equation (38) and can be solved in the same manner as equation (4) by defimng the matrix ξ as [P' ^• C,Δ ^• P' ^• C] and d e vector K as [S,M].

Altematively and/or in addition, an adjusted matrix P, P^, can be calculated using previously determined values of S and M . Using equation (22), d e adjusted matrix P is given by:

P_adj.i = P(ω; + S + (co; - ω_c) M) . (42) when the sum S + (ω - ω_c) M is large and by equation (23):

P_adj = P + P' ^• C + Δ ^• P' ^• C. (43) when S + (ω - ω.) M is small. In eitiier case, P_ιdj can be used in place of P in any of equations (1), (38), or (41) to obtain a best fit value of C that is conected for botii types of shift. If eitiier equation (38) or (41) are used, new values for S and M are also obtained.

In a particularly prefened embodiment of die present invention, d e foregoing methods are applied to analytical absorbance spectroscopy. In this embodiment of the invention, equation (1) is the well known Beer-Lambert law:

A_ob,(λ) = E(λ) ^• C, (44) where A^λ) is the absorbance spectrum measured as a function of the wavelength λ,

E(λ) is the matrix of wavelength-dependent extinction coefficients for die absorbents, and

C is the concentration of the absorbents. A is a vector whose "m" elements A; are the absorbencies at "m" discrete wavelengths λ;, E is an "m x n" matrix whose elements E,, are the extinction coefficients of component "j" at wavelength λ_j, and C is a vector, each of whose "n" elements C_j is the concentration of absorbant "j".

Hence, it is seen that in this particularly prefened embodiment, the following conespondences exist: ω = λ (45a)

_ dA

Y' = (45c) dλ

P = E (45d)

5P

P' (45e) dλ

10. ΔΔ,j == ((λλ-j -- λλ_cc)) (450

These definitions are then used in equations (1) through (43) to measure d e magnitude of the experimental enor and calculate a value of C tiiat is conected for enor, as described above.

This embodiment is particularly useful, for example, in analytical UV-VIS-IR

15 absorbance spectrophotometry (e.g. , as conducted wid d e Ciba-Corning Diagnostics 800 Series Co-oximeters) used to determine die concentrations of blood components in blood samples. The major components of blood that will ordinarily be used in d e present methods are reduced hemoglobin (HHb), oxyhemoglobin (O₂Hb), carboxy hemoglobin (COHb), metiiemoglobin (MetHb), sulfhemoglobin (SHb), and lipid. The wavelengtii- 0 dependent extinction coefficients for hemoglobin-based blood components for use in E can be measured directiy (preferably on a highly calibrated spectrophotometer) or obtained from the literature. E.g. , Zijlstra et al., Clin. Chem. 37(9), 1633-1638 (1991). The lipid spectrum can be measured from intravenous fat emulsion (e.g., the commercially available lipid product intralipid) in an aqueous dispersion of about 10% by weight. In a 5 particularly prefened aspect of the invention, the inventive method is used to conect for instrumental wavelength drift observed in UV-VIS-IR absorbance spectroscopy.

In another embodiment, the foregoing methods are employed to conect for experimental enor in column chromatography. In this embodiment, die independent variable ω is equal to die elution time, t. The spectrum Y_ob, is the magnitude of die signal for detecting die presence of sample components in die eluate (e.g., absorbance, refractive index). The shape of the elution profile measured under standard and controlled conditions provides die elements of P. In this embodiment, a shift in elution time, dt, is equivalent to die shift in the independent variable, dω . Such a shift may arise due to deviations from the standard conditions of such parameters as die flow rate, solvent strength, and column temperature. The change in elution time can be modeled as a combination of scalar shift and linear shift, as described in equations (37)-(43) and associated text. For example, a change in column flow rate affects the elution time for unretained components and will change the elution time for retained components in approximate proportion to their initial elution times less the elution time for the unretained components. Therefore, all peaks will be shifted by a fixed delay plus one proportional to d e elution time difference.

All of d e foregoing mathematical manipulations can be conducted using standard software packages, such as MATHCAD (MathSoft, Cambridge, MA).

In all of d e methods of die present invention, a sample spectrum is generated from which the sample component concentrations can be determined. The sample component concentrations are then determined by employing one or more of the methods described previously.

The apparatus according to die invention incorporate the methods according to the invention to more accurately determine the concentrations of sample components. Accordingly, the apparams according to the invention comprises a means for generating a spectrum (as defined hereinabove) of an analytical sample, a means for detecting the spectrum, a means for recording the spectrum, and a means for manipulating die spectrum according to any of die methods described herein. Myriad means for generating, detecting, and recording a spectrum exist and are well known to those skilled in die art. E.g. , Hobart H. Willard et al., Instrumental Methods of Analysis (7th Ed., Wadsworth Pub. Co., Belmont, CA, 1988). The means for manipulating die spectrum comprises any computer means that can run software embodying one or more of the foregoing methods for determining sample component concentrations conected for experimental enor from the measured spectrum. Of course, the practitioner will appreciate tiiat die apparatus need not be an integrated unit. In a prefened embodiment, me apparatus is a spectrophotometer capable of generating a sample spectrum in the UV, VIS, or IR regions of the electromagnetic spectrum. In another prefened embodiment, the apparams comprises column chromatography equipment.

The following examples are provided to illustrate certain embodiments of die invention and are not intended, nor should tiiey be construed, to limit the invention in any manner.

EXAMPLES All mathematical manipulations described herein were performed using the MATHCAD software by MathSoft (Cambridge, MA). Ciba-Corning Diagnostic proprietary extinction coefficient matrices were used.

Example 1 Estimating Wavelength Shift Using the Derivative ofthe Absorbance Spectrum The absorbance spectrum of eleven blood samples was measured at a resolution of 1 nm and d e fractions of HHb, O₂Hb, COHb and MetHb calculated using a least squares solution of equation (42). The results are as follows:

0.5 -0.5 0.2 1.2 0.4 0 0.8 0.4 2.2 0.2 -0.4

93.3 100.4 98.7 97.4 98.4 100 98 98.4 97.1 94.4 99.6

(a)

5.1 -0.2 -0.1 0 0 -0.3 -0.1 0.2 0.6 4.8 0.4

1.2 0.4 1.1 1.3 1.2 0.4 1.3 1 0.1 0.6 0.3

where each column is one sample and d e rows are die fractional concentrations of HHb, O₂Hb, COHb and MetHb, respectively. The data were shifted by 0.1 nm by fitting die measured spectmm widi a cubic spline and shifting die fitted spectmm. The fractional concentrations were then determined by solving equation (42) for C using the least squares method. The result was:

-0.3 -1.3 -0.5 0.4 -0.4 -0.8 0.1 -0.4 1.5 -0.5 -1.2

93.7 101 99.3 98 99 100.6 98.6 98.9 97.6 94.9 100.2

(b)

6.1 0.7 0.8 1 0.9 0.6 0.8 1.1 1.5 5.8 1.4

0.5 -0.3 0.4 0.7 0.5 -0.4 0.6 0.3 -0.6 -0.1 -0.4 The difference between the original data (a) and the shifted data (b) was determined to be:

-0.8 -0.8 -0.8 -0.8 -0.8 -0.8 -0.8 -0.8 -0.8 -0.8 -0.8

0.4 0.6 0.6 0.5 0.6 0.6 0.6 0.6 0.5 0.4 0.6

(c)

1 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 1 1

0.7 -0.7 -0.7 -0.7 -0.7 -0.7 -0.7 -0.7 -0.7 -0.7 -0.8

As can be seen from matrix (c), the relative change in fractional concentrations due to a frequency shift as small as 0.1 nm can be quite pronounced, particularly for components that are present in relatively small concentrations.

Using the derivatives of the absorption data spectra, the shifts and concentrations were calculated on a sample by sample basis using equation (22). The results for the shift were:

(0.11 0.04 0.07 0.14 0.09 0.15 0.11 0.11 0.17 0.09 0.05) , (d) and results for the concentrations were:

The difference between matrices (a) and (e) was calculated to be:

0.1 -0.5 -0.2 0.3 -0.1 0.4 0.1 0.1 0.5 0 -0.4

-0.1 0.3 0.2 -0.2 0 -0.3 -0.1 -0.1 -0.4 0 0.3

(f)

-0.1 0.6 0.3 -0.3 0.1 -0.4 -0.1 -0.1 -0.6 0.1 0.5

0.1 -0.4 -0.2 0.3 -0.1 0.4 0.1 0.1 0.5 0 -0.4

By comparing matrices (c) and (f) it is seen that die differences from d e original, unshifted spectrum are much smaller when one accounts for die shift. The average shift from matrix (d) is 0.104 nm. Note that even though the individual estimates of S can vary by as much as 50%, the average value is exceptionally good. Using die average shift of 0.104 nm to compensate for the shift in the spectmm of each sample, the fractional concentrations were calculated using equation (24):

0.5 -0.5 0.3 1.3 0.4 0 0.9 0.4 2.3 0.3 -0.3

93.2 100.4 98.7 97.4 98.4 100 98 98.3 97.1 94.4 99.6

(g)

5 -0.3 -0.1 0 -0.1 -0.4 -0.1 0.2 0.6 4.7 0.4

1.2 0.4 1.1 1.4 1.3 0.4 1.3 1 0.1 0.6 0.3

and die difference between d e original fractional concentrations (matrix (a)) and d e fractional concentrations calculated using the average over 11 values of die shift, S, (matrix (h)), was determined to be:

Given that the average value of die shift was a very good estimate of the acmal shift, it is not suφrising that the calculated estimates of the concentrations matched die concentrations of the unshifted data so well.

Example 2 Estimating Wavelength Shift Using the Derivative of the Extinction Coefficient Matrix Using the same data tiiat resulted in matrix (a) in Example 1 and shifting the spectmm by 0.1 nm, we compensated for die shift by die method of calculating the derivative of the extinction coefficient matrix as described in equations (18)-(27) and associated text using estimated values of the component concentrations obtained by solving equation (42). The estimated fractional concentrations were determined to be:

which differs from the original concentrations of matrix (a) as follows:

0.1 -0.5 -0.2 0.3 -0.1 0.4 0.1 0.1 0.5 0 -0.4

-0.1 0.4 0.2 -0.2 0 -0.3 -0.1 0 -0.4 0 0.3

0)

-0.1 0.5 0.3 -0.3 0.1 -0.4 -0.1 -0.1 -0.6 0.1 0.4

0.1 -0.4 -0.2 0.3 -0.1 0.4 0.1 0.1 0.5 0 -0.3

The wavelength shifts, S, were determined to be:

(0.1 0.04 0.07 0.13 0.09 0.15 0.11 0.11 0.16 0.09 0.05 0.03) . (k) The average shift, S , is 0.097 nm.

Using the average estimated shift, S , and solving for the concentrations C using equation (74) yielded die following concentrations:

0.5 -0.6 0.2 1.3 0.4 -0.1 0.9 0.4 2.3 0.3 -0.4

93.2 100.4 98.7 97.3 98.4 100 97.9 98.4 97 94.3 99.6

(1)

5.1 -0.2 -0.1 0 0 -0.3 -0.2 0.3 0.6 4.8 0.5

1.2 0.4 1.1 1.4 1.3 0.3 1.3 1 0.1 0.6 0.3

which resulted in the following differences from the original concentrations in matrix (a):

Example 3

Averaging S Over the Previous Eight Measurements

Using the data from Example 1 and focusing on the ninth sample, one can average over the first eight samples to obtain a value of S . The original fractional values for the ninth sample (obtained from the unshifted spectmm) were:

and die fractional concentrations after shifting the spectmm by 0.1 nm were:

The average shift of the first eight samples was 0.096. The fractional concentrations and wavelengdi shift were calculated using die derivatives of die extinction coefficients. The fractional concentrations obtained were:

and die calculated wavelengdi shift was 0.06. The shift estimated for the first sample was 0.1. Continuing to average the shift over the last 8 samples, the new average shift S is given by

0.096 x 8 - 0.1 0.06

= 0.103

7 8

Using this revised figure for S , the adjusted value for the extinction coefficient matrix,

E_adJ, was calculated and used to obtain a revised set of fractional concentrations:

which, for this sample, differs insignificantly from the previous estimate.

Claims

We claim:

1. An improved mediod of determining the concentrations of one or more components of an analytical sample from an observed spectrum that can be estimated by

Y (ω) = P(ω) ^• C, wherein Y^, is a vector whose "m" elements are the magnitudes of die observed spectrum at each value of an independent variable ω, C is the vector whose "n" elements are the estimated concentrations of "n" components diat contribute to the measured spectrum, and

P is a "m x n" matrix whose elements are the magnitudes of die contribution to the spectrum of each of "n" components at each of the "m" values of die independent variable ω, wherein the method comprises generating a sample spectrum from which the concentrations of the sample components can be determined and determining the sample

• component concentrations from the spectrum, the improvement comprising conecting for experimental enor by modeling die experimental enor as "r" types of enors given by the product ξ ^• K, where K is a vector whose "r" elements are the magnitudes of each of the "r" types of experimental enors and ξ is an "m x r" matrix whose elements are the relative enors at each value of ω for each type of experimental enor, adding die product ξ ^• K to the estimated spectmm as

Y_ob,(ω) = P(ω) ^• C + ξ ^• K and solving for the best fit values of C and K, wherein "n" and "r" are integers each greater or equal to 1 and "m" is an integer at least "n+ r".

2. The method according to claim 1, wherein the best fit values of C and K are determined by least squares analysis.

3. The mediod according to claim 2 wherein r is 1 and die experimental enor is modeled as a shift in ω by amount dω, and ξ = Y' = (dY_(Λ)^/(dω).

4. The method according to claim 2 wherein r is 1 and die experimental enor is modeled as a shift in ω by an amount dω, ξ = P' ^• C, and

5. An improved method of determining the concentrations of one or more components of an analytical sample whose observed spectmm can be estimated by d e equation

Y ω) = P(ω) ^• C wherein Y^ is a vector whose "m" elements are me magnitudes of die observed spectmm at each value of an independent variable ω, C is the vector whose "n" elements are the estimated concentrations of "n" components that contribute to the measured spectrum, and P is a "m x n" matrix whose elements are the magnitudes of die contribution to die spectrum of each of "n" components at each of die "m" values of die independent variable ω," wherein the method comprises generating a sample spectrum from which the sample component concentrations can be determined and determining the sample component concentrations from the spectrum, die improvement comprising conecting for experimental enor by modeling the experimental enor as a shift of the spectrum by an amount dω , estimating a shift, dω , calculating an adjusted spectrum, Y^, using eidier the equation:

or:

and determining the concentrations, C, by solving either the equation: Y_adj = P ^• C or:

Y^ = P ^• C + ξ ^• K. for the best fit value of C, where ξ takes the form ξ = Y' = (dY_ob (dω) or ξ = P' ^• C, where

F' - dω .

"m" and "n" are intergers, "n" is greater or equal 1 and "m" is at least "n."

6. A mediod according to claim 5 wherein dω is die weighted average of "k" previous values of dω, wherein k is an integer greater or equal to 1, and dω is determined from the equation dω = w^{τ •} dω^* , where w^τ is the transpose of a vector w of length "k" whose elements are the relative weights to be applied to each of the previous "k" values of dω, dω* is a vector of length "k" whose elements are previously determined values of dω, and w satisfies die equation:

Σ ^w, = ^{i •}

7. The method according to claim 6, wherein each element of w is 1/k, or:

or

where "a" is any real number greater than 1 and "i" is an integer from 1 to "k."

8. An improved mediod of determining die concentrations of one or more components of an analytical sample whose observed spectmm can be estimated by d e equation Y_obJω) = P(ω) ^• C wherein Y^ is a vector whose "m" elements are the magnitudes of the observed spectrum at each value of an independent variable ω , C is the vector whose "n" elements are the estimated concentrations of "n" components that contribute to tiie measured spectrum, and P is a "m x n" matrix whose elements are the magnitudes of die contribution to the spectrum of each of "n" components at each of the "m" values of the independent variable ω, die improvement comprising conecting for experimental enor by modeling the experimental enor as a shift in ω by an amount dω , estimating a shift, dω . calculating an adjusted P matrix, P^, using either the equation:

P_adj = P_obs(ω + ~ ) or:

and determining the concentrations, C, by solving either the equation:

or: Y_ob,(ω) = P^ω) ^• C + ξ ^• K. for the best fit value of C, where ξ takes the foπn ξ = Y' = (dY^ dω) or ξ = P' C, where dω and "n" is greater or equal to 1 and "m" is at least "n."

9. A metiiod according to claim 8 wherein the weighted average of dω , d ω , is die weighted average of k previous values of dω, wherein k is greater or equal to 1, and determined from the equation

3ω = w^τ • dω^* , wherein w^τ is the transpose of a vector w of length "k" whose elements are the relative to be applied to each of die previous k values of dω, wherein "k" is an integer greater or equal to 1, dω* is a vector of length "k" whose elements are the previously determined values of dω, and w satisfies the equation:

∑ w_f = l ^■ i=l 10. The method according to claim 9, wherein each element of w is 1/k, or:

wherein "a" is any real number greater than 1, and "i" is an integer from 1 to "k."

11. The method according to any of claims 3, 4, 6, or 9, wherein the shift in ω is modeled as constant across the entire spectrum and equal to a scalar dω = S.

12. The method according to any of claims 3, 4, 6, or 9, wherein the shift in ω is modeled as dα>j = (G>_J - ω^M, wherein dOj is the shift at the i^ώ value of ω, G>_J is the \^Λ value of ω, ω_c is a constant value of ω, and M, which can be any real number, is the magnitude of die enor.

13. The method according to claim 11, wherein Y is an absorbance spectrum of the sample, P is the extinction coefficient matrix for the absorbing components of the sample, and ω is d e frequency at which the absorbance is measured.

14. The method accordmg to claim 12, wherein Y is an absorbance spectmm of the sample, P is the extinction coefficient matrix for the absorbing components of the sample, and ω is d e frequency at which the absorbance is measured.

15. An apparatus for determining the concentration of components in a sample comprising a means for generating a spectmm from the sample; a means for detecting the spectmm; a means for recording the spectrum; and a means for determining d e concentrations of sample components from the spectrum according to d e method of claim 13.

16. An apparatus for determining d e concentration of components in a sample comprising a means for generating a spectrum from the sample; a means for detecting d e spectrum; a means for recording die spectrum; and a means for determining d e concentrations of sample components from the spectrum according to the mediod of claims 14.

17. The apparatus according to claim 15, wherein the apparatus comprises a spectrophotometer capable of generating a sample spectrum in the UV, VIS, or IR regions of the electromagnetic spectrum.

18. The apparatus according to claim 16, wherein the apparatus comprises a spectrophotometer capable of generating a sample spectrum in the UV, VIS, or IR regions of the electromagnetic spectrum.