Apresentao do PowerPoint

1
APLICAÇÕES DA QUIMIOMETRIA EM DIFERENTES ÁREAS DA
QUÍMICA   Márcia M. C. Ferreira Laboratório de
Quimiometria Teórica e Aplicada Instituto de
Química UNICAMP Email marcia_at_iqm.unicamp.br URL
http//lqta.iqm.unicamp.br
2
Quantitative analysis and classification of AFM
images of human hair   S. P. GURDEN, V. F.
MONTEIRO, E. LONGO J. Microsc., in press.
3

• This work aims to build a computation algorithm
  capable of analysing an AFM image and calculating
  a set of parameters which describe as fully as
  possible the cuticular structure.
• These descriptors are used for automatic
  classification of hair samples according to
  factors such as distance from the root-end and
  hair treatment.

The central cortex of a hair fibre is surrounded
by thin cellular sheets, known as cuticles, which
overlap each other from root to tip. These
cuticles fulfil a number of useful roles
including protection from physical and chemical
insult and a tendency to maintain the hair in a
clean and disentangled state
4
Two types of black, Caucasian human hair samples
were used (a) untreated and (a) bleached. . Each
hair fibre had a thickness of approximately 65?m
and a length of approximately 20cm. The AFM
images were measured using a Digital Instruments
NanoScope IIIa instrument under atmospheric
conditions at 25?C using a loading force of
3.6nN.
AFM images of Caucasian hair
bleached hair near the distal end
untreated hair near the root end
5
Original hair image
Background surface calculated during the
planification step
Hair image after planification
6
The quantitative analysis and classification of
AFM hair images is carried out by characterizing
the hair surface using descriptors which
summarize the important characteristics of the
cuticular structure. Step height Tilt Backtilt
Layer spacing Face
distance(AB) Top distance(BC) Fit error
Roughness Fitability Cuticle density
(cuticles per mm). .
total length
B
backtilt
step height
layer spacing
tilt
cuticle surface
A
C
underlying cuticle
7
Cuticular descriptors calculated for the example
image shown before.
Schematic of PLS-DA
dummy variables
descriptors
0 1 1 1 1 0 etc.
X
Y
samples
samples
PLS
The data was autoscaled before building the PLS
model. From 38 samples, the model was built using
36 (2 outliers). Of the 36 samples, 31 were
classified correctly, giving a success rate of
86.
descriptors
R
P
R
T
PLS loadings
PLS scores
samples
8
The samples (?) are closely clustered. They are
the most homogeneous in terms of cuticle
structure, as the hair surface has not been
greatly damaged by physical or chemical stress.
They have similar descriptor values regardless of
which particular image is analysed. The
untreated/distal end (?) and bleached/root end
(?) samples also form fairly well-defined groups
(exception are the two untreated/distal end
samples and the misclassified samples).

• untreated/distal end
• ? bleached/root end
• untreated/root end
• ? bleached/distal end.

The bleached/distal end samples (?) are very
scattered, indicating that representative
sampling of this group is more difficult. As PC1
describes negative contributions from cuticle
density and fitability, it is logical that the
samples damaged by cosmetic treatment and
long-term physical stress (far from the root end)
have the highest PC1 scores and so, have low
cuticle density and poor fitability.
9
Root and distal end samples In general, the root
end samples (? and ?) have negative component 2
scores and the distal end samples (? and ?) have
positive component 2 scores. This shows that at
the root end of the hair, where the cuticles are
more abundant, the degree of tilt and step height
of the cuticles is higher as may be expected. At
the distal end, where the cuticles are less
abundant, the cuticles lie flatter and have a
lower degree of backtilt due to physical wear.
Distal end cuticles also have a less uniform
pattern, also due to the effect of prolonged
physical stress which chips away at the cuticle
ends, leaving an irregular cuticle edge. Effect
of bleaching For the samples measured at the root
end, the bleached samples (?) lie further to the
right that the untreated samples (?). This
indicates that one effect of bleaching is to
remove the cuticles which protect the central
hair cortex, thus making the hair less resistant
to breakage or splitting. The bleached samples
also have lower component 2 scores, symptomatic
of cuticle detachment which also leaves the hair
in a weaker condition, more vulnerable to
subsequent damage. For the samples measured at
the distal end (untreated, ? bleached, ?), the
removal of cuticle layers is even more
pronounced, showing that bleaching of already
vulnerable distal end hair can lead to complete
removal of the cuticular layer in some cases,
exposing the underlying central cortex.
10
Quantitative Determination of Epoxidized Soybean
Oil Using Near-Infrared Spectroscopy and
Multivariate Calibration       Thais F.
Parreira Henrique J. S. Sales, and Wanderson
B. de Almeida  Henkel S/A Indústrias
Químicas Appl. Spectrosc., 56, 1607-1614 (2002).

11
Soybean oil is a triglyceride which typically
contains 14 stearic, 23 oleic, 55 linoleic and
8 linolenic acid. Three of them are unsaturated
acids oleic (181), linoleic (182) and
linolenic (183). Chemical modification on
commercial available soybean oil such as
epoxidation can enhance its properties
(reactivity) for industrial applications.
12
The epoxidized soybean oil (ESO) is extensively
used in plastic industry as plasticizer (at
levels ranging from 0.1 to 27), to increase
flexibility and as stabilizer to minimize
decomposition in polyvinyl chloride (PVC)
products.
PVC degradation caused by sunlight eliminating
HCl
Reaction of oxirane ring with HCl inhibiting the
degradation process.
13

• To follow the soybean oil epoxidation process, it
  is necessary to quantify some analytes related to
  the products quality
• The epoxide index (E.I.) is directly related to
  the stabilizer feature of the product ? the
  higher the epoxide content, the more efficient
  the additive as thermal stabilizer.
• The iodine content (I.I.) is an indicator to
  the amount of unsaturations present in the
  epoxidized soybean oil.
• The percentage of water ( of Water), results
  from washing of the final product. Its
  concentration must be minimal since water can
  cause degradation of the epoxide group.

Iodine index determination ? halogenation of
double bonds.
Degradation of epoxide groups by water
14
NIR absorbance spectra were recorded from 9300
cm-1 to 4500 cm-1 with a 2 cm-1 increment, using
a BOMEM MB160 FTIR spectrophotometer.
ORIGINAL DATA SET (2400 VARIABLES)
Box car average
REDUCED DATA SET (160 VARIABLES)
Generic recorded spectrum
SELECTION A (Spectral difference)
SELECTION B CORRELOGRAM
SELECTION C (Loadings Regression vector)
EXTERNAL VALIDATION (using the best model)
15
7050-7290, 5190-5310 cm-1
WATER
7110-7230, 5130-5310 cm-1
6930-7350, 4500-5910 cm-1
16
PLS Models
a Variable selection by the correlogram applied
to Selection B for cutoff 0.80. b External
prediction using PLS model with 14 variables
selected in Selection A. c SEP (standard error
of prediction).
FIGURA
17
Experimental, estimated, and residual values for
water for external set. PLS model from
Selection A (3 LVs)
aRPD std(exp.)/std(residuals). bRER
range(exp.)/std(residuals).
18
CONCLUSIONS

• ? The use of NIRS combined with multivariate
  regression is a feasible
• alternative to the widely established
  techniques, especially in industrial
• processes.
• ? Using simple and intuitive variable selection
  methods, such as
• loadings/regression vector analysis and the
  correlogram, the number
• of variables can be significantly reduced
  without impairing the model
• quality.
• ? The statistical parameters used, RPD and RER,
  indicated that NIRS
• determination was accurate for E.I. (RDP26.0
  and RER80.7) and
• fairly good for water and I.I.
• ? From the results obtained, it can be concluded
  that the proposed
• methodology is appropriate for monitoring the
  epoxidation of soybean
• oil and to evaluate the additives quality in
  the industrial process,
• where time, effort and money are crucial.

19
Chemometric and Molecular Graphics and Modeling
Study on Bacterial b-Lactam Efflux Mechanism by
Multidrug Resistance AcrB Pump Rudolf
Kiralj
20
ABSTRACT

• The primary purposes of this work
• To establish relationships between activity
  expressed as log of minimal inhibitor
  concentration (pMIC) elevated by three strains of
  Salmonella typhimurium (HN891, SH7616, SH5014),
  and calculated descriptors for 16 penicillins and
  cephalosporins at neutral pH.
• To visualize pump drug molecular recognition
  mechanism, using crystal structure of AcrB
  transporter from Escherichia coli.
• ? These results can aid in explaining bacterial
  drug efflux mechanism, and design of novel
  b-lactams which would not be excreted from
  bacterial cells.

21
INTRODUCTION
? Antibiotics are characterized by their chemical
composition and mode of action.   ? Penicillins
and cephalosporins have the cell wall as target
for their action.   ? b-lactam antibiotics are
the most used antibacterial inhibitors of the
Penicillin- Binding-Proteins (PBPs), which
are responsible for the construction and
maintenance of bacterial cell wall. ? There are
different mechanisms by which bacteria exhibit
resistance to antibiotics  
1-  Bacteria produce b-lactamases which hydrolyze
the b -lactam antibiotic ring before
their binding to PBPs.   2-  Bacteria change
their permeability to the drug (passive membrane
transport).   3-  Bacteria develop a
structurally altered PBP that is still able to
perform its metabolic function, but less
affected by the drug.   4-  Bacteria change
their express transport system that actively pump
the drug to the outer cellular
environment.
22

• The major mechanism of MDR in bacteria is the
  pump drug efflux. In general
• this is accomplished by the presence of
  AcrAB-TolC efflux systems, which
• are responsible for the unidirectional
  pumping of a wide variety of lipophilic
• and amphiphilic compounds out of the cell.
• FACTORS THAT INFLUENCES THE MULTI DRUG EFFLUX
  RATE
• Pumps number
• Substrate concentration
• pH
• Highly charged residues
• Substrate charged groups
• MDR PUMPS consist of 3 components
• 1- a resistance-nodulation-cell division
  transporter AcrB (trimeric)
• 2- an outer membrane channel protein of the
  family TolC (trimeric)

23
General
AcrAB-TolC bacterial pump. S. Murakami et al.,
Nature, 419 (2002) 587.
24
METHODOLOGY
MICs for bacterial strains ? Mass concentration
MICs (from literature) for 16 b-lactams effluxed
by bacterial strains S. typhimurium SH5014
(parent strain), SH7616 (an acr mutant) and HN891
(an overproducer of the Acr pump). Drugs
Modeling ? Molecular structures were refined or
modeled by Spartan Pro using atomic coordinates
from PPSD, CSD or 2D formula. Conformational
search was done by Montecarlo method and the most
stable conformers were optimized by the
semiempirical method PM3. Lipophilicity
Parameters ? logarithm of the octanol-water
partition coefficient logKOW was from Nikaido et
al and several others were calculated using
different approaches. wC, Sf ? are the number
fraction and surface fraction of hydrophobic
carbon atoms, respectively. Other molecular
descriptors geometrical, electronic and Hydrogen
bond molecular properties were calculated using
2D or 3D geometry of the antibiotics.
25
Molecules
16 antibiotics (penicillins and cephalosporins)
as AcrB substrates
26
Correlation of pMICS
Comparison among three pMICs pHN891 and pSH5014
are highly correlated (right). pSH7616 shows
different trend (left). The three bacterial
strains are not distinguished when excreting
highly charged antibiotics.
27
Chemometrics of pMICs
PCA and HCA were performed using only pMIcs data.
b-Lactams were classified as good, moderately
good to poor, and bad AcrB substrates. Clustering
of b-lactams with respect to the number of
charged groups NCH and hydrophobic surface
fraction Sf is visible.
28
Chemometrics of lipophilicity descriptors
PCA (left) and HCA (right) analysis of 9
lipophilicity descriptors logarithm of the
octanol-water partition coefficient (logP)
calculated by various methods, surface fraction
(Sf) and number fraction (wC) of hydrophobic
carbons. Two clusters and two isolated logPs are
visible. The lipophilicity descriptors do not
contain the same information (82.8 of the
variance contained in PC1 PC2).
29
Lipophilicity pMIC relationships
An example of lipophilicity activity nonlinear
relationship. Log Kow was linearized by
GlogKOW exp(logKOW 1.1)2 Other
transformations SlogPs (logPs)2 SlogKWIN
(logKWIN)2.
30
PLS regression models for pMICs
It is visible that the best PLS models are
obtained when all types of parameters are used
lipophilic, electronic and hydrogen bonding.
31
Experimentala and Predicted pMICSH5014b
Except for sample 9, exp-cal differences are
smaller than 10.
aH. Nikaido et al., J. Bacteriol., 180 (1998)
4686. bMIC are in mols per liter.
32
AcrB crystal structure
S. Murakami et al., Nature, Nature 419 (2002)
587.
Science 300 (2003) 976.
Crystal structure of the AcrB trimer determined
by X-ray diffraction protein without (left) and
with a ligand (right). Three distinctive units
are visible TolC docking domains, Pore domains
and Transmembrane domains. The system of cavities
and channels for drug efflux can be also noted
the three vestibules, the large central cavity,
the narrow pore, and the cone-like funnel.
33
Left efflux mechanism by AcrAB-TolC pump. Right
AcrB trimer in two views.
34
Efflux mechanism and the inner structure of the
AcrB protein.
35
The vestibule structure
Left Electrostatic potential of the pore anf the
transmembrane domains.
The vestibules projection has functional surface
through which the drug can pass without
difficulty. This area is called BRAMLA, due to
its resemblance with the map of Brazil (BRAzil
Map-Like Area). The upper third of BRAMLA is
surrounded by hydrophilic and the other two
thirds by hydrophobic residues of the AcrB.
36
The vestibule-drug interactions
Left Schematic representation of drug-vestibule
stereolectronic complementarity that was deduced
from similarity of the 16 antibiotic structures
and importance of lipophilic, electronic and
hydrogen bonding molecular parameters. Molecular
recognition is obvious, and it can be weaken or
enhanced by the nature of R and R1 side chains.
Right 3D docking of nafcillin (1) to the
vestibule. Interactions between hidrophilic AcrB
residues (in rectangles) and nafcillin polar
groups are visible.
37
2D docking of selected AcrB substrates to the
BRAMLA area, using maximum and minimum (right)
stereoelectronic fit approach for some
antibiotics. It can be noticed that the
antibiotic molecules differ in how well then can
fit sterically and electronically to the
vestibule. These fittings correspond to
biological activities for the presented
antibiotics.
38
The pore structure
The structure of the pore channell (left figures)
and the pore recognition site (right figures)
viewed perpendicularly to or along the three-fold
axis of the AcrB protein. The pore channel
consists of three short a-helices and three
random coils. The pore recognition site contains
highly hydrophobic (yellow) and hydrophilic (red
or pink) residues these residues are selective
with respect to drugs due to hydrophobic, polar
and hydrogen bond interactions.
39
The pore-drug interactions
Some drugs docked to the pore recognition
site. Lipophilic drugs enter the pore channel
easier than hydrophilic ones due to 1) weaker
intermolecular interactions 2) more favourable
drug-pore recognition. These conclusions, based
on 3D docking of the presented drugs, are in
agreement to chemometric results.
40
Substrate 14 bound to a portion of a pore from a
protomer (left) and its electrostatic potential
at molecular surface in free and bound state
(right). There are four drug pore hydrogen
bonds involving residues Ala100, Ala103 and
Gln104. This illustrates why hydrogen bonding and
electronic descriptors are important in the PLS
models.
41
CONCLUSIONS
PLS models of good quality were obtained using
lipophilic, electronic and hydrogen bond
descriptors for 16 b-lactams. Proposed efflux
mechanism based on chemometrics and molecular
graphics and modeling methods 1) a drug
molecule comes from periplasmic space and
interacts with a vestibule through a mechanism of
molecular recognition ? large and highly
hydrophilic molecules hardly enter the vestibule
and come to the central cavity of AcrB
protein. 2) a drug molecule from the central
cavity comes to the pore recognition site and
through a mechanism of molecular recognition
enters the pore channel ? again large and highly
hydrophilic molecules hardly enter the pore
channel to be excreted from the cell.
42
A METHODOLOGY FOR IDENTIFICATION OF ATMOSPHERIC
POLLUTION SOURCES             Edilton de S.
Barcellos
43

• This work introduces a methodology to identify
  the principal emission pollution sources in the
  Região Metropolitana de São Paulo.
• The analysis covered the primary pollutants CO,
  NO, NO2 and CH4, and the secondary one O3.
• The data (kindly provided by the Sanitation
  Department of the State of São Paulo, CETESB),
  are time series consisting of concentrations
  measured hourly throughout the year of 1999 for
  each compound, in the site of P. D. Pedro II.
• To capture the systematic variations for each
  compound, the data was firstly arranged as
  matrices 24 (hours of the day) ? 365 (days of the
  year) and submitted to a Principal Component
  Analysis (PCA).
• To extract simultaneously the daily and weekly
  systematic variations, the data was rearranged in
  a multiway structure (24 hours of the day ? 7
  days a week ? 52 weeks of the year) and the
  Tucker model was applied.

44
METHODOLOGY

• ?Outliers
• Outliers were identified from a visual inspection
  of the original data matrix and considered as
  missing values in the case their values were ten
  times larger than the mean value for that
  pollutant.
• ?Missing Data
• The matrices containing missing data were
  subjected to the mdpca routine of the PLS-Toolbox
  (Eigenvector Research, Inc.) (MATLAB software
  version 5.1) was used.
• ?Data Preprocessing
• For individual pollutants no preprocessing was
  applied.
• In order to minimize the local (hourly) sudden
  variations, when the analysis was carried out for
  more than one pollutant simultaneously, the data
  set was standardized by the mean concentration of
  the pollutant in question, i.e., for a given
  pollutant, each entry value was divided by the
  mean value taken from all the data relative to
  that compound.
• ?Constraints
• Non-negativity was the only constraint used.
• The multiway analyses were performed by using the
  N-way Toolbox 1.02 for MATLAB (http//newton.foods
  ci.kvl.dk/Matlab/nwaytoolbox).

45
Construção da matriz 365 (dias) ? 24 (horas) para
o CO e sua decomposição em matrizes de scores
e de loadings.
Threeway representation for CO (array X) and its
decomposition by PARAFAC model into
component matrices A, B and C.
46
Construção da estrutura multimodo mostrando a
disposição dos dias, horas e semanas nas
matrizes.
47
Carbon Monoxide PCA showing daily systematic
variation throughout the year
48
CO, NO and NO2 have a profile that is not similar
to the profile of CH4. This result indicates that
there are at least two emission sources. One of
them is a road traffic source. The profile for
CH4 in this site may be due to industries that
uses natural gas, and sludge treatment plants,
among others less important.
49
The major sources of ozone in the troposphere are
the photochemical reactions and the necessary
conditions are a source of carbon (CO and/or
hydrocarbons), nitrogen oxides (NO?, NO2?), and
sunlight . Carbon Monoxide Cycle The chain of
chemical reactions that occurs in the troposphere
has the real starting point with the
photochemical decomposition of ozone, resulting
in molecules and oxygen atoms in the excited
O3 h? ??? O2 O? O? H2O ??? 2
HO? Depending on the concentration of
nitrogen oxides (especially NO?) in the
environment, the reaction can follow two
different paths. Low NOx conc. High NOx
conc. CO HO? ?? CO2 H? CO HO? ?? CO2
H? H? O2 M ?? HO2? M H? O2 M ??
HO2? M HO2? O3 ?? HO? 2 O2 HO2? NO? ??
HO? NO2? Net CO O3 ?? CO2 O2
NO2? h? ?? NO? O O O2 M ?? O3 M
Net CO 2 O2 ?? CO2 O3
50
TUCKER MODEL
(a) Daily emission source profile for the primary
vehicular pollutants (PC1), for other primary
pollutants (PC2), and for O3 (PC3) (b) Weekly
systematic variations for the primary vehicular
emissions (PC1), other primary sources (PC2) and
O3 (PC3) (c) Annual variations for the primary
pollutants (PC1) and for the secondary ones
(PC2) (d) Traffic density profile in the RMSP.
51
CONCLUSIONS

• ?The models were able to identify the different
  primary sources and distinguish the primary from
  secondary pollutants.
• It was possible to explain the weekly source
  profiles, and the weekly profile for the
  secondary pollutant, O3 through reactions
  occurring in both atmospheric cycles namely the
  carbon monoxide cycle and methane cycle.
• The proposed models have pointed out the
  importance in reducing not only NOx, but also CO
  and CH4 emissions from local anthropogenic
  sources for controlling local levels of ozone and
  global warming.

52
Lapachol e derivados de 1,4 naftoquinonas em
carsinosarcoma W-256   Subramanian S.,Trsic
Milan (USP SC) Structural Chemistry 9, 47
(1998).
53
Lapachol é uma naftoquinona extraída do caule de
certas bignoniáceas da Ásia e América do Sul. Uma
destas plantas o nosso IPÊ ROXO. É um pó de cor
amarela intensa.
54
O Lapachol e alguns de seus derivados foram
testados com bons resultados experimentais em
tumores como o walker 256 (W-256) carcinoma e
várias pesquisas estão hoje sendo feitas. Neste
exemplo, a estrutura do lapachol e vários
derivados de 1,4 naftoquinonas são usados para
investigar a relação entre parâmetros estruturais
e a atividade biológica usando KNN e mais tarde
usando SIMCA. O conjunto de dados consiste de 25
compostos extraídos da referência citada acima e
são classificados como ativos e inativos. Os
descritores (variáveis) são os coeficientes da
função de onda, , do orbital molecular de mais
alta energia, dos seguintes átomos de carbono b,
c, m, n, o, p, q, s, t, u e foram obtidos de
cálculos semiempíricos usando o método PM3.
Representação 3D do HOMO
Representação 3D do LUMO
(c) INATIVA (XXIII)
(a) INATIVA (XXIII)
(b) ATIVA (III)
(d) ATIVA (III)
55
A análise de componentes principais, nos dados
autoescalados, mostra que PC1 discrimina os
compostos ativos dos inativos. Do gráfico de
escores e da tabela dos loadings, ve-se que os
compostos ativos têm uma alta contribuição das
variáveis p-u (loadings negativos) indicando uma
alta densidade eletrônica na dupla ligação da
cadeia lateral e grupos terminais. Os compostos
inativos têm alta contribuição dos átomos b-n
(loadings positivos). Provavelmente a quinona é
capaz de participar de alguma reação de
oxi-redução como agente redutor onde os elétrons
?p da dupla ligação agem como doadores de
elétrons.
56
SIMCA 2 PCs for each class
57
LQTA Laboratório de Quimiometria Teórica e
Aplicada