The next section d escrib es th e re s u lts of th e egg white sep aratio n s and th e re s u lts of th e identifications using th e peaks found by deconvolution. The param eters for th e peak functions found by deconvolution a re listed in table A2.11 and th ese param eters were in te g rate d to produce th e peak area data listed below in tab le 3.16. The deconvolution of each separation was carried out to a d ifferen t d eg ree of accu racy in term s of th e p recisen ess to which th e overall peak sh ap es were found - ie., th e positions of th e peak maxima, s ta r ts and e n d s as well as th e peak areas. (A discussion of th e effectiv en ess of th e deconvolution tech n iq u e in determ ining th e c o rre c t peak shapes fo r th e egg white chrom atogram s is given in section 2.3.10.)
Clearly th e ability of th e deconvolution tech n iq u e to determ ine accu rately th e peak areas will affect th e su b se q u e n t ability of th e fuzzy logic identification tech n iq u e to identify acc u ra tely th e peaks. The information determ ined by th e identification ro u tin e may, as well as being used to identify th e components in th e separation, be used to comment on th e effectiv en ess of th e deconvolution algorithm . A low mean match c riterio n between a tria l deconvoluted chromatogram and th e refe ren c e chromatogram data indicating a poor re s u lt fo r both th e matching algorithm and deconvolution algorithm .
T a b le 3.16 D eco n v o lu ted E g g W hite C h ro m ato gram P e a k A rea D ata ( p e a k d a ta . O b je c tiv e f u n c tio n < -1 8 3 .6 ). trial Egg White Chromatogram Number Peak Area ( mV . mL ) 1 2 3 Total (1+2+3) Actual Total Area 1 118.0 45.5 118.3 281.8 285.4 2 103.7 47.4 116.3 267.4 274.1 3 100.8 70.0 102.4 273.2 262.3 4 175.0 51.5 110.9 337.2 339.8 5 100.0 45.8 114.6 260.4 263.3 6 48.9 39.8 134.7 223.5 223.6
Calculated by in te g ratin g th e whole chromatogram
Such combinations of poor re s u lts may be due to a num ber of reasons. The composition of th e mixture sep arated may change - e ith e r th e amounts or th e identities of th e components. This may lead to a change in th e total chromatogram area. In th is situ atio n it would be reasonable to expect a low overall match c riterio n since one or more of th e areas of th e peak functions comprising th e chromatogram will be d ifferen t. If th e total area of th e tria l chromatogram is th e same as th e refe ren c e chromatogram b u t with only a low mean match criterio n th en one may assum e th a t th e deconvolution is inaccurate. Poor deconvolution may be because th e c o n strain ts used in th e deconvolution p rocess were in ap p ro p riate (see section 2.2.4.1) , th e model used to d escrib e th e peak elution profiles is in ap p ro p riate (these should be checked f ir s t - see section 2.3.8.1), or th e chromatogram is not su itab le for analysis by deconvolution (see section 2.4).
The re s u lts of th e identification of th e deconvoluted peaks a re listed in tab les contained in appendix A3. The firs t, second, and fifth chromatogram pose no serio u s problems to th e matching algorithm and th e components a re co rre ctly matched to th e ir tem plate component in th e refe ren c e data set. However th re e of th e chrom atogram s listed in table 3.16 (num bers 3, 4 and 6) contain peak a re as which differ significantly from areas in th e o th er chrom atogram s. The reaso n s fo r th e se d ifferences a re discussed below to g e th e r with comments on th e su b seq u e n t effect
upon th e identification of th e peaks.
The th ird egg white chromatogram contains a c e n tra l peak which has an a re a approxim ately 25 mV.mL above th e o th er reliable m easurem ents of th e c e n tra l peak. The total a re a for th is chromatogram is not sig n ifican tly d ifferen t from th e total a re a s of most of th e o th er chrom atogram s and th is implies th a t if th e area of th e c en tral peak (and hence th a t of th e th ird peak) was c o rre ctly determ ined th en th e identification of th e peaks would be more certain . (The reaso n s for th e in accu rate deconvolution a re d iscu ssed in sections 2.3.10 and 2.2.6.) The c en tral peak in th is chromatogram is matched with th e c en tral peak in th e re fe re n c e d ata set b u t only with an overall match c riterio n of 0.56. Both of th e o th e r peaks a re matched with much h ig h er overall match c rite ria (in excess of 0.9). The fo u rth chromatogram has a to tal a re a well in excess of th e o th e rs and th e a re a of th e f ir s t peak is much la rg e r th an in o th er chrom atogram s (even as a proportion of th e to tal chromatogram). The peak shape obtained by deconvolution is not an u n -reaso n ab le one and a la rg e r peak are a would be expected since th e peak h eig h t is h ig h er th a n in o th e r egg white chrom atogram s. This implies th a t th is p a rticu la r chrom atogram may contain e ith e r a d ifferen t proportion of th e same components as in th e o th er chrom atogram s or th e chromatogram contains e x tra o r d ifferen t com ponents. In o rd e r to determ ine which of th e se situ atio n s had o ccu rred re q u ire d f u r th e r detailed analysis of th e column e lu e n t - p a rtic u la rly components not contained within th e molecular w eight ra n g e th a t was analysed by SDS PAGE (see section 2.3.5). In addition to th ese points in d icatin g th a t th e m ixture may not be of exactly th e same as in th e o th e r egg white chrom atograms is th e ap p earan ce of e x tra peaks in th e chrom atogram . The identification p ro cess gives a v e ry poor match for th e f ir s t peak as expected given its larg e a re a b u t good m atches fo r th e second and th ird peaks.
The six th egg white chrom atogram is su b je c t to se v ere o v erlap p in g of peaks and con seq u en tly all peaks a re v e ry in accu rately determ ined and th u s all peaks a re poorly matched - p a rticu la rly th e f ir s t and second peaks. In fa c t all peaks a re matched with significantly lower overall match c rite ria .
3.5 Conclusions
The matching algorithm d escribed in th is c h ap ter has been shown to successfully identify component peaks within chrom atograms using a referen ce or tem plate chromatogram.
• Determination of th e tech n iq u es limitations and process situations leading to failure.
The algorithm has been shown to be applicable to several ty p es of chrom atograms and its limitations illu stra te d using simulated chrom atographic data. The limitations occur in th e main where peak c h a ra c te ristic s (namely peak a re a and elution ord er) in th e refe ren c e and or tria l data se t are similar. This is p a rticu larly tr u e when sim ilarity exists between more th an one ch arac teristic. The fau lt which is most likely to cause th e tech n iq u e to produce an in co rrect identification is th e ap p earan ce of a peak of an additional peak in th e tria l chromatogram which has similar identification c h a ra c te ristic s to a peak in th e refe ren c e data set. Not only does th is affect th e identification of neighbouring peaks b u t all peaks in th e chromatogram and may cause an in co rre ct identification. These problems with sim ilarity of peak c h a ra c te ristic s a re exactly th e same as those encountered by a human analysing such a problem. However, by using th e fuzzy logic tech n iq u e th is allows a more q u an titativ e analysis to be carried o u t resu ltin g in a more co n sisten t level identification being achieved.
The methods for determ ining how similar c h a ra c te ristic s need to be before problems occur in th e matching pro ced u re have been described.
• Application to chrom atogram s req u irin g deconvolution. Chromatograms containing sev erely overlapped peaks were also analysed successfully. The su ccess of th e analysis is dependent upon th e d eg ree of overlap and th e relativ e h eig h ts of th e peaks within th e chromatogram. An im portant factor with th e su ccessfu l identification of th e peaks is th e accu rate deconvolution of th e chromatogram to obtain th e peak functions d escribing th e individual component elution profiles (see C hapter 2). This is re q u ire d so th a t th e amount
of a given component may be a cc u rately determ ined (ie. th e peak area). Providing th a t th is information is obtainable th en th e peaks within th e chromatogram may be matched accurately with peaks in th e re fe re n c e chromatogram.
• Analysis of se v ere overlap:
More sev ere overlapping has also been considered where th e degree of overlap is such th a t no valley ex ists between th e two peaks and in certain circum stances peaks have been successfully identified. The u se of th e tech n iq u e to identify v e ry heavily overlapped peaks (such as th o se obtained by poor deconvolution) and th e ir th e composition has been investigated using simulated data.
The following c h ap ter of th is th e sis co n sid ers th e combined u se of deconvolution data and peak identification tech n iq u es in th e control of fraction selection for th e optim isation of a chrom atographic process, th e th ird process listed on fig u re 1.8.
I
%I
0
1.0 0.0 XElution Rank ( Dimensionless )
Figure 3.1 The elution order membership function for a reference peak. The measured rank is X.
1.0 M M B m atch criterion m easurem ent (a) threshold
difference in m easurem ents (b)
Figure 3.2 (a) The m atch criterion is obtained from the overlap o f m em bership functions
(b) A m atch is considered to be good if the m atch ctrierion is above the threshold.
60
50 40 30 20 10 0 0 2 4 6 8 10 12 14 16Elution Volume (Arbitraiy Units)
Figure 33 The area data contained in table 3.1 may be represented ly this chromatogram
I
I
1.2 0.8 0.6 0.4 0.2 10 20 30 40 50Peak Area ( Arbitraiy Units )
60
Figure 3.4 Peak Area Membership Functions for the Reference Data Set listed in table 3.1. Hie maxima of the membership functions correspond to the peak areas in table 3.1.
1.0- 0.8 — 0.4
Ô
0.2- 0.0 ▲ Aa
-T--- 1--- r 2 3 4Trial Peak Number ( - ) □ reference 1 A reference4 A reference2 ■ references V references — - threshold Figure 3.5
The overall match criterion for the matching of the data in table 3.2 with the data in table 3.6. Each datat set shows the matching of each trial peak with a reference peak. This figure shows a successful identification since only one data set (ie. only one reference peak) at each trial peak co-ordinate is above the threshold.
0 .9 - O. SÀ 0 .7 - § 0.6 O ûû 0.5 â 0.4 0 .3 - 0.2- 0.1 - T 2 3 4 5 6
Actual Trial Peak Number ( - )
reference4 a reference5
threshold
Figure 3.6
The overall match criteria for the fourth and fifth reference peaks in table 3.7 compared with each reference peak in table 3.1. This indi cates that a problem occurs with the fifth trial peak since both the fourth and fifth reference peaks have overall match criteria abpve the threshold.
C hapter 4. Control.
4 Control
Once th e individual component elution profiles have been obtained (see section 2.1) and th e components matched to a tem plate chromatogram (see section 3.1), it is th en possible to c a rry out control actions based on th e information gained from th e se processes. The aim of th e work in th is th e sis is to control p rep a rativ e sep aratio n s where th e chrom atography is c arried out to produce m aterial of a defined r a th e r th an to e ffect a high d eg ree of resolution so th a t th e composition of a m ixture may be determ ined. Hence control actions will be in stitu te d to maintain th e p u rity of th e p ro d u ct and th e p ro d u ctiv ity (ie. th e amount of m aterial produced per u n it time or separation) of m aterial produced. In su ch a process situation th e issu e of d etecto r sa tu ratio n may be critical, since as described in section 2.3.1 th e analysis of chrom atograms with d etector satu ratio n re q u ire s th e u se of an on-line or a t-lin e analytical system such as HPLC. The chrom atogram s produced from each on-line analysis may th en provide th e source of data on which to base control decisions. Analysis of th e information produced by th e analytical system will, as for th e process chromatogram, re q u ire some p rio r knowledge of th e n a tu re of th e chromatogram produced by th e a t-lin e system . If th e chosen analytical technique is a colum n-based chrom atographic method th en th e information obtained will be in th e form of a chromatogram with (provided th e size of m aterial loaded is determ ined co rrectly ) lin ear d etecto r response. The n a tu re of th is chromatogram may be d iffe re n t from th e p rep a rativ e chrom atograms analysed in section 2.3.8. Knowledge of which components correspond to th e n e a re st neighbour(s) on e ith e r side of th e p ro d u ct is req u ire d since th e elution o rd e r may not be th e same as in th e process chromatogram as e ith e r a d ifferen t ty p e of chrom atography may be used or displacem ent effects may occur a t a p ro cess scale due to high loadings used which may affect th e elution o rd e r of some components. That w ithstanding th e same generic ap p ro ach as d escrib ed in th is th e s is will apply to deconvolution, peak identification, and control.
Control actions which may be tak en a re th e selection of th e optimum fraction based on a defined se t of p ro d u ctiv ity and p u rity c rite ria and secondly th e re-optim isation of th e sep aratio n should it not be possible to obtain a fraction with th e d esired p ro d u ctiv ity and p u rity . Since deconvolution and fuzzy logic identification re q u ire th a t th e whole chromatogram be available for analysis on-line control u sin g th e se tech n iq u es is not possible. The following actions may be ta k e n at-lin e:
(i) The calculation of th e position of th e optimum fractio n for su b seq u e n t sep aratio n s. If th e fraction selection action is to be applied on su b se q u e n t sep aratio n s th en it is essen tial th a t th e variation in overall