Size Exclusion Chromatography - American Chemical Society

---

Figure 3b depicts these same rules in the form of a hierarchical decision tree. The first determination is wheth­ er a carbonyl peak is present between 1765-1660 c m - 1 and is of strong in­ tensity (intensities range from 7 to 10). If a peak is present in the correct range of positions and intensities, the probability of the compound being an aldehyde is set to 0.05. If no appropri­ ate peak is present, the aldehyde tree is finished. If a carbonyl peak is pres­ ent, queries about the C-H stretching region are then posed. "True" re­ sponses to these questions would lead to a further increase in the probability index. Rather than continuing to lump all aldehydes together as a single class, three subclasses have been created: saturated aldehydes (no alpha-beta unsaturation), unsaturated aldehydes (in at least the alpha-beta position), and aldehydes containing an alphaelectronegative group. A scientist would attempt to make this distinc­ tion; PAIRS should do so as well. A real aldehyde would yield sufficient true decision tree responses to result in a probability index greater than 1.0. PAIRS has been tested on many spectra of varying complexity. In one study a total of 180 entries were drawn from the bulletins of the International Center of Information on Antibiotics.

Many of the 56 classes involved were interpreted very successfully by PAIRS. Four classes—acetal, ketal, methylene, and nonaromatic olefin— were handled poorly by PAIRS. Each of these functionalities is difficult to interpret accurately using infrared spectroscopy. PAIRS made a total of 10,080 predictions (180 spectra X 56 functionalities) and was correct 88% of the time. Eliminating the four most difficult classes led to predictions that were more than 90% correct. As software tools that expedite the development of expert systems be­ come more prevalent, the availability and usefulness of these systems in an­ alytical chemistry and spectroscopy will increase markedly. Note: PAIRS, including the knowl­ edge base, is available for unrestricted distribution as Program 426, Quantum Chemistry Program Exchange, Indi­ ana University, Bloomington, Ind. 47401.

L a w r e n c e Livermore National Laboratory Livermore, Calif. 9 4 5 5 0 Contributor:

Carla

M.

Wong

Problem area. Lawrence Liver­ more National Laboratory (LLNL) is implementing an expert system for in­

strument control of a totally comput­ erized triple-quadrupole mass spec­ trometer (TQMS). This system is an ideal candidate for self-adaptive, or knowledge-based, control for several reasons. The spectrometer is a very complex instrument with more than 30 computer-controlled parameters, and it produces vast amounts of data. Most important, the skill required for tuning or optimizing operating condi­ tions is the kind of knowledge that can be represented as procedures and rules (2). TQMS is a multistage instrument consisting of an ion source, a quadrupole mass analyzer, an rf-only quadrupole collision gas chamber, a second quadrupole mass analyzer, and an ion detector (3). Figure 4 shows the basic operational modes. It also shows the opportunities for the varied experi­ mental data that can be collected and the increased operational complexity inherent to such systems. The prob­ lem is how to optimize for the most significant data, as opposed to merely collecting the most data. To do this one must encode into a knowledge base a tuning procedure for the TQMS that includes heuristics to describe what will eventually become a selfadaptive, feedback control process for real-time optimization of the data ac­ quisition procedure.

Size Exclusion Chromatography Methodology and Characterization of Polymers and Related Materials Theodore Provder,

CONTENTS D i r i i c . e Lh'omatography Modeling • Gel Permeation Chromatography iGPCi Moaelin.i • Supo'cnt-'al Fluid Chromatograph · Automated Data Anaiy&ie «&' GPC · SEC Us­ ui'.) Narrow anr< Broad Muleculi: Wfnqht UiM'iliiituri · SEC ot Polyethylenes · GPC Correction for I m p e d e d Resolution · Mnlp. ul.ir WrtiqM Sepa'atit-n and SEC Col.jTin D'spers on · Fuzes Statistical Mettions V Testing Identity «· Polymer MohA u'ar W.-irjht DistriDutif ι e Pnlymor Based HignF t V . i e i c ^ CiPC Columns · GPC Separation .)' Small Moleruies · Hiqh-Perrormnno< Hn)l ι Spend fiPC · Deiittinurn Oxide Use m Acu"™]-: S E C · Hic/i-PirformaniM GPC oi Pr|^iutlw!»nti turephtruriate) · Dearaaatior ·.« / i > · , H qti Mulucdl.tr Weight Pclyrmrs in ur-C · h.tfh-Lh".cienc> G P C Or Analysis ot Oi.qonr ··, arm Smalt Mnloi ulas e Petroleum Cmde a id Distillates Analysis By GPC · SEC of Poivothyli