Pattern recognition develops and applies algorithms that
recognize patterns in data. These techniques have important applications
in character recognition, speech analysis, image analysis, clinical
diagnostics, person identification, machine diagnostics,
and industrial process supervision.
Many chemistry problems can be solved with pattern recognition techniques,
such as: recognizing the provenance of agricultural products
(olive oil, wine, potatoes, honey, etc.) based on composition or spectra;
structural elucidation from spectra; identifying mutagens or carcinogens
from molecular structure; classification of aqueous pollutants based
on their mechanism of action; discriminating chemical compounds based
on their odor; classification of chemicals in inhibitors and non-inhibitors
for a certain drug target.
We now introduce some basic notions of pattern recognition.
A pattern (object) is any item (chemical compound, material, spectrum, physical object,
chemical reaction, industrial process) whose important characteristics form a
set of descriptors. A descriptor is a variable (usually numerical) that
characterizes an object. A descriptor can be any experimentally measured or
theoretically computed quantity that describes the structure of a pattern:
spectra and composition for chemicals, agricultural products, materials,
biological samples; graph descriptors and topological indices; indices derived
from the molecular geometry and quantum calculations; industrial process
parameters; chemical reaction variables; microarray gene expression data; mass
spectrometry data for proteomics.
The major hypothesis is that the descriptors capture
some important characteristics
of the pattern, and then a mathematical function (machine learning algorithm)
can generate a mapping (relationship) between the descriptor space and the
property. Another hypothesis is that similar objects (objects that are close in
the descriptor space) have similar properties. A wide range of pattern
recognition algorithms are currently used to solve chemical problems: linear
discriminant analysis, principal component analysis, partial least squares
(PLS), artificial neural networks, multiple linear regression (MLR), principal
component regression, k-nearest neighbors (k-NN), evolutionary
algorithms embedded into machine learning procedures, support vector machines.
An n-dimensional pattern (object)
x has n coordinates, x=(x1,
x2, …, xn), where each
xi is a real number, xi∈R
for i = 1, 2, …, n. Each pattern xj
belongs to a class yj∈{-1, +1}. Consider a
training set T of m patterns together with their classes,
T={(x1, y1),
(x2, y2), …,
(xm, ym)}. Consider a dot product
space S, in which the patterns x are embedded, x1,
x2, …, xm∈S.
Any hyperplane in the space S can be written as
The dot product w•x is defined by:
A training set of patterns is linearly separable if
there exists at least one
linear classifier defined by the pair (w, b) which correctly
classifies all training patterns (Figure 1). This linear classifier is
represented by the hyperplane H (w•x+b=0) and defines
a region for class +1 patterns (w•x+b>0) and
another region for class -1 patterns (w•x+b<0).
|
|
Figure 1. Linear classifier defined by the hyperplane H (w•x+b=0).
|
After training, the classifier is ready to predict
the class membership for new patterns, different from those
used in training. The class of a pattern xk is
determined with the equation:
Therefore, the classification of new patterns depends only
on the sign of the expression w•x+b.
|
