The performance of the heurlstlc evolvmg latent proJectIons (HELP) method 1s mvestlgated for peak purity control m hqmd chromatography Hrlth photodlode array detectlon HELP and evolvmg factor analysis performed equally well, both techmques could detect less than 1% of a spectrally slmdar lmpunty even for small chromatographic separations In certam cases, the two techmques may be hmlted by Instrumental and expertmental dlfflcultles Keywords Llquld chromatography, Heurlstlc evolvmg latent proJectIons, Peak punty control, Evolvmg factor analysis Hyphenated techniques such as liquid chromatography with photodlode array detection (LC-DAD) are becoming increasingly important for the analysis of multlcomponent systems In prmaple, a single analysis provides enough data to determine the number of species present in a mtiure and, by means of self-modelhng curve resolution [l-5], to resolve the data mto the spectra and chromatograms of the pure constituents Assessment of peak purity m LC-DAD by such methods should therefore be possible In practice, however, the determination of the number of chemical species IS not that simple Assessment of the rank of the bilinear data by a stat&l-Correspondence to D L Massart, Pharmaceutical Institute, VrlJe Umversltelt Brussel, Laarbeeklaan 103, B-1090 Brussels (Belgmm) cal test tends to overestimate the number of underlying species as shown by Gerrrtsen et al [6] Instrumental and experimental dlfflcultles may lead to artefacts when applying latent vanable methods on LC-DAD data Evolving factor analysis (EFA) for peak purity control m LC-DAD [4,7-81 can be limited by artefacts due to nonzero baselines, the DAD's scan time, a nonlinear cahbratlon curve and heteroscedastlclty
[9,10] Although these sources of errors may lead to senous problems, using the fared size window EFA (FSW) m combination with appropriate data pretreatment less than 1% of a spectrally similar isomer could be detected under a chromatographic peak [ill This paper studies the performance of the recently developed heuristic evolvmg latent projections (HELP) method [12,13] to detect a small