Neural net system for analyzing chromatographic peaks

• US 5,121,443 A
• Filed: 09/21/1989
• Issued: 06/09/1992
• Est. Priority Date: 04/25/1989
• Status: Expired due to Fees

• Current Assignee: Thermo Instrument Systems, Inc. (Thermo Fisher Scientific Incorporated)
• Original Assignee: Spectra-Physics Incorporated (MKS Instruments Incorporated)
• Inventors: Tomlinson, Barrett L.
• Primary Examiner(s): Mancuso, Joseph
• Application Number:
  US07/410,552
• Time in Patent Office: 992 Days
• Field of Search: 382/29, 382/17, 382/15, 382/14, 364/497, 364/498, 364/496, 364/513, 210/656, 73/23.22, 73/23.23, 73/23.35, 73/23.36
• US Class Current: 382/156
• CPC Class Codes: G01N 30/8624 Detection of slopes or peaks; baseline correction
• CPC Class Codes: G01N 30/8631 Peaks
• CPC Class Codes: G01N 30/8641 Baseline

Abstract

A two-step process for characterization of peaks in a chromatogram is disclosed. Firstly, data corresponding to each peak or pair of peaks in the chromatogram is identified. A unique filter apparatus locates extrema of the curvature of the chromatographic data and a data file is generated containing characteristics of the extrema. A pattern recognition apparatus analyzes the characteristics of the located extrema and classifies the peak or peak combination represented by the data in the file as one peak or peak combination in a set of resolved peaks and selected combinations of resolved peaks. A portion of the chromatographic data, corresponding to the peak or peak combination identified by the pattern recognition apparatus, is identified. This portion includes the signal for said peak and the signal for the baseline upon which the peak is superimposed. In the second step, data for a peak or a peak combination identified as described above is processed and a set of characterizing parameters for the peak or the first peak in the peak combination is generated without a prior baseline correction to the data. The peak data including the baseline level upon which the peak is superimposed is analyzed using one of lookup-tables, neural nets, curve fitting, or combinations thereof. These characterization processes, using information about the peak crest and the peak inflection points, determines a set of characterizing parameters and a baseline estimate that best fit the identified data. The peak characterization is not biased by a prior baseline correction.

Claims

1. A neural net system for characterizing a peak superimposed on a baseline comprising:

a source of data signals representing said peak superimposed on said baseline;

an input layer comprising a plurality of input units operatively coupled to said source of data signals, each input unit generating an output signal in response to an input signal from respective ones of said data signals;

at least one hidden layer comprising a plurality of hidden units operatively coupled to said input units, each hidden unit generating an output signal in response to a plurality of output signals from said input units which is a weighted function of said output signals from said input units; and

an output layer comprising a plurality of output units operatively coupled to said plurality of hidden units, each output unit generating an output signal in response to a plurality of output signals from said hidden units wherein

the output signals from an output unit represent best estimates of a set of parameters which characterize said peak, said set of parameters from said output unit being determined without prior baseline correction to said data signals.

2. The system of claim 1 further comprising:

means, operatively coupled to said source of data signals and to said input units, for differentiating each of said data signals; and

means for providing said differentiated data signals as input signals to said input units.

3. The system of claim 2 wherein means for differentiating said data signals calculates the second derivative with respect to time of said data signals.

4. The system of claim 1 wherein each of said output unit output signals comprises a weighted function of said plurality of hidden unit output signals.

5. The system of claim 1 in which said output signals from said output units include at least one observable parameter comprising observed retention time.

6. The system of claim 1 in which said output signals from said output units algebraically transformed into exponentially modified gaussian parameters.

7. The system of claim 6 in which said exponentially modified gaussian parameters include at least δ

, the standard derivation of the gaussian function.

8. The system of claim 6 in which said exponentially modified gaussian parameters include at least τ

, the time constant of the exponential decay function.

9. The system of claim 6 in which said exponentially modified gaussian parameters include at least t_g, the time of maximum amplitude of Gaussian function.

10. The system of claim 6 in which said exponentially modified gaussian parameters include at least h, the peak vertical amplitude of the peak.

11. The system of claim 1 in which the amplitude of said data signals to said input layer are prescaled.

12. The system of claim 1 in which the width of said data signals to said input layer are prescaled.

13. The system of claim 1 in which said output signals from said output units include at least one observable parameter comprising observed peak height.

14. The system of claim 1 in which said output signals from said output units include at least one observable parameter comprising the square root of the second moment of the measured peak.

15. The system of claim 1 in which said output signals from said output units include at least one observable parameter comprising the ratio of the exponential decay constant to the gaussian peak width.