|
What...
Classifion is a freeware Windows application for chemometrics classification of
substances using their mass-spectra. The used multivariate analysis is
based on Principal Component Analysis with Mahalanobis Distance (PCA-MD)
and it is of one class classifier. Classifion has been
tested using mostly mass and optical spectra, but there is no limitation
to be used with any other kind of characteristic spectra. The only
twist is to find appropriate channel (bin) size, which in case of
mass-spec is naturally the mass number.
Where does it sit...
Almost every mass-spectrometry lab is equipped with spectral-library
search program with tens of thousands of spectra of pure compounds. The
problems begin when you have very complex substances or using mass-spec
methods different from those used in the spec-libraries. In some cases
you don't need to know what exactly is in the sample, but simply does it
belong to a class of samples you already measured. In all these cases
you need to use classification software. There are many classification
libraries in python or R, but they are general
purposed and require certain knowledge of the statistical methods used.
Here Classifion sits, offering you a classification with
specialized for mass-spectrometry friendly environment and automation allowing
you to classify without interpreting the mass-spectra, so you can
concentrate on your immediate work. Some understanding of underlying
statistical methods would be beneficial for you, but it is not required.
How does it work...
Classifion needs to be "trained" using several measured spectra per
substance. The "training" itself could be conducted automatically or
manually (supervised). The aim of the training is to extract substance
spectra specific information based on statistical characteristics, not the
interpretation of the mass-spectra. Mathematically speaking the extraction is down to
reducing the dimensionality of the spectra variable space (to each mass
corresponds one dimension). PCA is well-known such technique, which
offers other advantages as ordering the principal component by
"significance" (useful for noise reduction). Mahalanobis
distance will measure a distance between a new (unknown) point and the
cluster (the class distribution) in PCA space.
|
Classifion can train itself entirely on autopilot.
Just import your data, run Autopilot, sit back and enjoy the view. The
results will show you how successful the training was, so in case of
problem you should re-examine your data or do supervised training.
Classifion is one class classifier. Each training
correspond to one class and the unknown samples are tested against each
training individually. That approach is more convenient - adding new
class requires calculating only that class training and better from
precision prospective - each training contains and it is optimized for
information specific to that particular class. That makes the work with Classifion similar to the
spectral-library search software. Classifion can be
controlled remotely from another application (as COM server) combining
remote with user access (see Client.zip
example project).
What's new in v1.7
-
The optimization procedure
has been improved with calculating the classification thresholds
based on MDs histogram of the base group and the others against the
training.
-
A lot of user interface
improvements and some reorganization were implemented.
-
New utilities have been
added:
-
Import Data -
import external data files with varieties of options (Excel
style)
-
Spec-tree Builder -
generates a spec-tree/spec-groups files from appropriate
folder structure
-
External Datasets -
import a number of datasets from Classifion website
|