In Silico Spectra Lab Slide 1 In Silico Spectra Lab Explore & - - PowerPoint PPT Presentation

Slide 1

In Silico Spectra Lab

Slide 2

Libraries Management

(import spectra from different sources)

Libraries Management

(import spectra from different sources)

Explore & investigate unknown samples against spectra libraries Extract qualitative & quantitative information

(use of advanced mathematical methods)

Explore & investigate unknown samples against spectra libraries Extract qualitative & quantitative information

(use of advanced mathematical methods)

Build real & virtual calibrations Build real & virtual calibrations Deploy for monitoring & control @ production plant Develop & integrate custom process control algorithms & software Deploy for monitoring & control @ production plant Develop & integrate custom process control algorithms & software

In Silico Spectra Lab

Slide 3

In Silico Spectra Lab - software features and key points

Libraries Management

(import spectra from different sources)

Libraries Management

(import spectra from different sources)

Explore & investigate unknown samples against spectra libraries Extract qualitative & quantitative information

(use of advanced mathematical methods)

Explore & investigate unknown samples against spectra libraries Extract qualitative & quantitative information

(use of advanced mathematical methods)

Build real & virtual calibrations Build real & virtual calibrations

Slide 4

In Silico Spectra Lab – Libraries management

List of spectra List of spectra Description and formula Description and formula

Slide 5

In Silico Spectra Lab – explore & investigate …

Case setup

Unknown target spectrum has to be resolved (demixed, unmixed) quantitatively and qualitatively in a linear combination of component spectra. The user doesn’t know exactly which are the components but tipically several libraries of spectra are available; also the user has to be sure whether NIR technique can or cannot solve this problem.

Notes & observations

• Typical approach in R&D and method development problems.
• Very powerful in situations where the preparation of mixtures from pure components

spectra is impossible and/or chemometrics is not usable, i.e. : solid samples (minerals …).

• Often the objective is the detection and quantitative measurement of the presence of a

family of substances and not a specific molecule. A molecule present in the target may be unkown and therefore not present in the user libraries but its family is represented in one library (i.e. the presence of an unknown explosive substance falling into a known chemical family; the same is applicable for detection/suspection of the presence of new drugs …).

Slide 6

In Silico Spectra Lab – … explore & investigate

Solution

First, the user selects several spectra (components) from available libraries. These spectra undergo a clustering process which groups spectra in families (the clustering can be executed in total unsupervised or partially supervised mode). For each established family a synthetic representative spectrum is generated (the centroid of the cluster). The user, based on information contents and differences, chooses some or all the synthetic representative spectra as candidate components in the UnMix calculation.

The unmixing process finds the best combination of representative spectra and eliminates those

who are not present in the target The user can then choose a real spectrum for each representative and repeat the unmixing process with real spectra To help this process the “Mathematics” capabilities of the software allows the user to apply custom formulas to single spectrum and/or to custom combine more spectra to generate syntethic spectra.

Slide 7

In Silico Spectra Lab – … explore & investigate

Main advantages and key points

Competitors allows the un-mixing only for a retricted number of known components while InSilico Spectra Lab gives the user the possibility to use potentially all the availables spectra as demix candidate. When the user has such a freedom he/she has the tendency to use his/hers chemical knowledge when choosing the demix candidates. Often, expecially in NIR analysis, the spectra of different molecules can be similar. This leads to the results that several combination of different components can give rise to good unmix results, creating thus confusions and indetermination. To avoid this indetermination, a robust classification (clustering) algorithm is implemented. This algorithm uses pure numerical information contained in the spectra, avoiding the indetermination due to the use of chemical knowledge. The clustering algorthm itself, gives the possibility to invetigate the goodness of the NIR technique for the separation of molecules in classes/families as much as orthogonal among themselves. This leads to very robust UnMix calculations or to the results that the NIR information is not suitable to solve the user problems. Manies objective mathematical measures based on the raw spectra information is given in order to guide the user, guarantee reproducibility, avoid unreal results obtained by users which proceed in a kind of brute force trial-and-error way based on subjective adn aprioristic interpretation of NIR spectra.

Slide 8

In Silico Spectra Lab – … explore & investigate

… main advantages and key points

The clustering algorithm performs a hierarchic classification (dendrogram) based on the distances between the spectra in the library. The distance can be choosen as Euclidean distance, dot product or spectral correlation. if the user has previously defined the classes of the libraries, according to his knowledge about the characteristics of the samples in the library, the clustering algorithm reports how similar its classification has been in comparison to the one defined by the user. In other words, it reports some mathematical indices of the similarity between the calculated classification and the expected classification. the clustering algorithm creates, for each cluster, a synthetic spectrum that is the centroid of the

• cluster. This new sample can be added in the library and used as a new component.

Slide 9

In Silico Spectra Lab – explore & investigate Clustering

Slide 10

In Silico Spectra Lab – explore & investigate Clustering

Slide 11

In Silico Spectra Lab – explore & investigate … UnMix

!"

Slide 12

In Silico Spectra Lab – explore & investigate … Mathematics

Slide 13

In Silico Spectra Lab – virtual calibrations

Slide 14

In Silico Spectra Lab – quantitative Chemometrics …

Slide 15

In Silico Spectra Lab – … quantitative Chemometrics

Slide 16

In Silico Spectra Lab – … quantitative Chemometrics

Slide 17

In Silico Spectra Lab – Chemometrics Prediction