FEASABILITY OF YEAST AND BACTERIA IDENTIFICATION

USING UV-VIS-SWNIR DIFUSIVE

REFLECTANCE SPECTROSCOPY

J. S. Silva, R. C. Martins, A. A. Vicente and J. A. Teixeira

IBB - Institute for Biotechnology and BioEngineering, Universidade do Minho

Campus de Gualtar, 4710-057 Braga, Portugal

Keywords:

Yeast, bacteria, UV-VIS-SWNIR reﬂectance spectroscopy, Singular value decomposition, Classiﬁcation.

Abstract:

UV-VIS spectroscopy is a powerfull qualitative and quantitative technique used in analytical chemistry, which

gives information about electronic transitions of electrons in molecular orbitals. As in UV-VIS spectra there is

no direct information on characteristic organic groups, vibrational spectroscopy (e.g. infrared) has been pre-

ferred for biological applications. In this research, we try to use state-of-the-art ﬁber optics probes to obtain

UV-VIS-SWNIR diffusive reﬂectance measurements of yeasts and bacteria colonies on plate count agar in the

region of 200-1200nm; in order to discriminate the following microorganisms: i) yeasts: Saccharomyces cere-

visiae, Saccharomyces bayanus, Candida albicans, Yarrowia lipolytica; and ii) bacteria: Micrococcus luteus,

Pseudomonas ﬂuorescens, Escherichia coli, Bacillus cereus. Spectroscopy results show that UV-VIS-SWNIR

has great potential for identifying microorganisms on plate count agar. Scattering artifacts of both colonies

and plate count agar can be signiﬁcantly removed using a robust mean scattering algorithm, allowing also

better discriminations between the scores obtained by singular value decomposition. Hierarchical clustering

analysis of UV-VIS and VIS-SWNIR decomposed spectral scores lead to the conclusion that the use of VIS-

SWNIR light source produces higher discrimination ratios for all the studied microorganisms, presenting great

potential for developing biotechnology applications.

1 INTRODUCTION

Spectroscopy is a powerful tool for biological appli-

cations, being applicable to liquids, solutions, pastes,

powders, ﬁlms, ﬁbres, gases and surfaces, and mak-

ing possible to characterize proteins, peptides, lipids,

membranes, carbohydrates in pharmaceuticals, foods,

plants or animal tissues (Hammes, 2005).

One of the most popular method is Infrared Spec-

troscopy (IR), and was ﬁrstly applied to biological

materials in 1911 (Riddle et al., 1956). In the 1950s

and 1960s research IR spectroscopy began to be ap-

plied for microorganism differentiation, but this re-

search was abandoned due to the unsatisfactory re-

sults obtained with dispersive spectrometers (Dziuba

et al., 2007). These were ignored during 20 years,

until modern interferometric Fourier Transform Infra-

Red spectrometers (FT-IR) and statistical comput-

ing methodologies became available (Dziuba et al.,

2007).

Recent techniques using FT-IR allowed microbio-

logical characterization and the discrimination at level

of sorting better species and strains. Attenuated to-

tal reﬂection and IR micro-spectroscopy have been

associated to the discrimination and identiﬁcation of

strains according to taxonomic classiﬁcation, gram

+/- factor, or even susceptibility to antibiotics and

grown medium (Mariey et al., 2001).

FT-IR has also been used to identify lactic acid

bacteria strains (e.g. Lactobacillus, Lactococcus,

Leuconostoc, Pediococcus and Streptococcus(Dziuba

et al., 2007), and the rapid identiﬁcation of Acineto-

bacter species (Winder et al., 2004). An extensive

FT-IR spectroscopy database for the identiﬁcation

of bacteria from the two suborders Micrococcineae

and Corynebacterineae (Actinomycetales, Actinobac-

teria) as well as other morphologically similar gen-

era was established in 2002 by Helene Oberreuter and

its team (Oberreuter et al., 2002). Furthemore, FT-

IR was used for the ﬁrst time to determine the ra-

tios of different yeast species (Saccharomyces cere-

visiae, Hanseniaspora uvarum) and two yoghurt lac-

tic acid bacteria (Lactobacillus acidophilus, Strepto-

coccus salivarius ssp. thermophilus) in suspensions

S. Silva J., C. Martins R., A. Vicente A. and A. Teixeira J. (2008).

FEASABILITY OF YEAST AND BACTERIA IDENTIFICATION USING UV-VIS-SWNIR DIFUSIVE REFLECTANCE SPECTROSCOPY.

In Proceedings of the First International Conference on Bio-inspired Systems and Signal Processing, pages 25-32

DOI: 10.5220/0001062800250032

 SciTePress

Table 1: Studied microorganisms characteristics and experimental conditions: Gram factor, colony colour and shape and

integration time.

Integration Time (ms)

Microorganism Gram Colony color Shape Medium UV-VIS (ms) VIS-NIR (ms)

Saccharomyces cerevisiae na white spherical YPD 70 66

Saccharomyces bayanus na white spherical YPD 45 61

Candida albicans na white spherical YPD 70 66

Yarrowia lipolytica na white rod YPD 70 61

Micrococcus luteus - yellow spherical TSA 20 19

Pseudomonas ﬂuorescens - translucent rod MP 20 30

Escherichia coli - translucent rod LB 30 13

Bacillus cereus + opaque rod LB 36 61

of distilled water (Oberreuter et al., 2000).

Raman Spectroscopy has also shown great poten-

tial for microorganisms identiﬁcations in microscopy,

such as for Candida yeast strains and bacterial strains

such as Staphylococus, Enterococus and Echerichia

Coli) (Guibeta et al., ). Applications are also found

in oral hygiene, for the identiﬁcation of Streptococ-

cus mutants, S. sanguis, S.intermedius and S. oralis

(Berger and Zhu, 2003). Moreover, this technique

is currently used to identify baterials cells of Staphy-

lococcus under different cultivation conditions (Harz

et al., 2005) and single yeast cells (Rch et al., 2005).

Fluorescence Spectroscopy (FS) is one of the most

important spectroscopic techniques in molecular bi-

ology, and consequently can also be used to microbi-

ological identiﬁcation. FS applications can be found

on the differentiation of yeast and bacteria, by their

intrinsic ﬂuorescence to UV excitation (Bhatta et al.,

2005).

UV-VIS-SWNIR spectroscopy is one of the most

widely used techniques in analytical chemistry, but it

has almost not been used for microorganism identi-

ﬁcation. This is perhaps attributed to the fact that

UV-VIS spectroscopy records transmitions between

electron energy levels from molecular orbitals, in-

stead of vibrational or structural oscillation of molec-

ular groups as in the infrared region. It is widely ac-

cepted that vibrational spectroscopy is more adequate

for organic chemistry measurement than transitional

spectroscopy. Nevertheless an asset of this technique

has never been done use for microbiological identiﬁ-

cation.

Electronic transitions in the UV-VIS region de-

pend upon the energy involved. For any molecular

bound (sharing a pair of electrons), orbitals are a mix-

ture of two contributing orbitals σ and π, with cor-

responding anti-bounding orbitals σ

∗

and π

∗

, respec-

tively. Some chemical bounds present characteristic

orbital conditions, ordered by higher to lower order

energy transitions: i) alkanes (σ → σ

∗

; 150nm); ii)

carbonyls (σ → π

∗

; 170nm); iii) unsaturated com-

pounds (π → π

∗

; 180nm); iv) molecular bounds to

O, N, S and halogens (n → σ

∗

; 190nm); and v) car-

bonyls (n → π

∗

; 300nm). As most UV-VIS spec-

trometers yield a minimum wavelength of 200nm, this

technique has been considered to provide lower infor-

mation in terms of functional groups when compared

to IR, being the spectral differences mostly attributed

to conjugated π → π

∗

transitions and n → π

∗

tran-

sitions (Perkauparus et al., 1994).

Only recording π → π

∗

and n → π

∗

tran-

sitions above the 200nm is however not totally a

handicap. Many organic molecules present con-

jugated unsaturated and carbonyls bounds, such as

aminoacids, phospholipids, free fatty acids, phe-

nols and ﬂavonoids, peroxides, peptides and pro-

teins, sugars and their polymers absorbance in these

bands. Furthermore, many biological molecules

present chromophore groups, which increase the ab-

sortion in the UV-VIS region, such as: nitro, nitroso,

azo, azo-amino, azoxy, carbonyl and thiocarbonyl,

which can be used to identify microorganisms.

UV-VIS-SWNIR has some advantages to FTIR

for microbiological identiﬁcation in plate count agar.

The lower wavelength turns this radiation attractive

due to the lower penetration, being easier to mon-

itor surfaces than NIR or MIR radiation. Further-

more, state-of-the-art ﬁber-optics miniature UV-VIS-

SWNIR are today affordable for mobile applications

such as identiﬁcation of microorganisms in surfaces,

using spectroscopy may became feasible in a near

future. Although UV-VIS retrieves only molecular

spectroscopy information, today’s equipments also

include high frequency vibrational spectroscopy in

the SWNIR region, giving important information on

water, fats and proteins which may be used to discrim-

inate between microorganisms.

In this research was tried to discriminate both

Yeasts and Bacteria of comonly used in microbi-

ology laboratories: i) yeasts: Saccharomyces cere-

visiae, Saccharomyces bayanus, Candida albicans,

Yarrowia lipolytica, ii) bacteria: Micrococcus luteus,

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

Figure 1: Yeast and bacteria growth media: (a) Tryptic Soy

Agar (TSA): Micrococcus luteus; (b) Pseudomonas Iso-

lation Agar (MP) (Pseudomonas ﬂuorescens); (c) Luria-

Bertani (LB): Escherichia coli, (d) LB: Bacillus cereus;

(e) YPD (Saccharomyces cerevisiae); (f) YPD (Saccha-

romyces bayanus); (g) YPD (Candida albicans); and (h)

YPD (Yarrowia lipolytica).

Pseudomonas ﬂuorescens, Escherichia coli, Bacillus

cereus; under plate count agar growth media.

The physical properties of the UV-VIS-SWNIR

spectra can also provide a great potential for the iden-

tiﬁcation of microorganism, using multivariate statis-

tical analysis and signal processing techniques. Mi-

crobes may not be directly identiﬁed by their main

colony chemical composition but rather by charac-

teristic metabolites produced under different growth

media. This is especially true for yeasts that ex-

hibit one of the most complex metabolisms in this

study. Therefore, not only the colony but changes

in the composition of the plate count agar in the sur-

roundings of each colony are expected to affect the

UV-VIS-SWNIR spectra in order to obtain signiﬁcant

discrimination between the different microorganisms

spectra. Therefore, the main objective of this research

work were to investigate the discrimination potential

of UV-VIS and VIS-NIR wavelengths to classify the

following microorganisms: i) yeasts: Saccharomyces

cerevisiae, Saccharomyces bayanus, Candida albi-

cans, Yarrowia lipolytica, ii) bacteria: Micrococcus

luteus, Pseudomonas ﬂuorescens, Escherichia coli,

Bacillus cereus; under plate count agar growth media.

2 MATERIALS AND METHODS

2.1 Sample Preparation

The microorganisms were obtained from the mi-

crobiological collection of the IBB - Institute for

Biotechnology and Bioengineering at the University

of Minho. The microorganisms were incubated un-

der aerobic conditions at 35

C during 72h. Micrococ-

cus luteus was cultivated on Difco Tryptic Soy Agar

(TSA) while Pseudomonas ﬂuorescens was grown on

Difco Pseudomonas Isolation Agar (MP). Escherichia

coli and Bacillus cereus were cultivated on Difco

Luria-Bertani Agar (LB). Yeast strains were grown on

Difco YPD Agar (YPD), at the same temperature and

time (Difco, 2005).

2.2 Spectroscopy

Yeast and bacteria UV-VIS-SWNIR spectroscopy

analysis was performed using the ﬁber optic spec-

trometer AvaSpec-2048-4-DT (2048 pixel, 200-

1100nm). Standart reﬂection UV-VIS and VIS-

SWNIR probes, models FCR-7UV200-2ME and

FCR-7IR200-2-ME (Avantes, 2007). A xenon and

halogen light sources, models AvaLight XE-2000

and AvaLight-Hal were used for UV-VIS and VIR-

SWNIR transmission measurements respectively; and

recorded using AvaSoft 6.0 (Avantes, 2007). Trans-

mission measurements were performed at the room

temperature of 18±2

C, and: (a) UV-VIS: the xenon

lamp was let to stabilize during 20 min; (b) VIS-NIR:

the tungsten lamp lamp was let to stabilize during 15

min. The dark spectra was recorded and measure-

ments were taken with linear and electric dark cor-

rection. Both light spectra were monitored by statisti-

cally assessing the reproducibility of the light source

with measurements of light during the several days

of the experiment. Fifteen spectra replicates were

recorded of UV-VIS and VIS-SWNIR measurement

of both plate count agar and microorganisms colonies

to study scattering effects. Futhermore, spectra were

obtained inside a box designed to isolate the environ-

mental light and maintain the probe at 90

angle with

the plate agar.

2.3 Spectral Analysis

2.3.1 Spectra Pre-processing

Table 1 presents the UV-VIS and VIS-NIR spectra ac-

quisition conditions. Experimental setup has shown

that it is impossible to use the same integration time

for the different microorganisms. Under these cir-

cumstances, all the collected spectra were normalized

norm

to remove this effect:

norm

= D

raw

max(x

)

(1)

where x

raw

is the original spectra, D

is the detec-

tor saturation value (14000 counts) and x

norm

the i’th

spectra normalized by its maximum value and resized

to the detector saturation.

Furthermore, as most plaque count agar growth

media are translucid, the signal recorded is in major-

ity the media information. To increase spectral vari-

FEASABILITY OF YEAST AND BACTERIA IDENTIFICATION USING UV-VIS-SWNIR DIFUSIVE

REFLECTANCE SPECTROSCOPY

400 500 600 700 800 900 1000

4 5 6 7 8 9

(a)

Wavelength (nm)

log(I)

400 500 600 700 800 900 1000

5 6 7 8 9

(b)

Wavelength (nm)

log(I)

400 600 800 1000

3 4 5 6 7 8 9

(c)

Wavelength (nm)

log(I)

400 500 600 700 800 900 1000

4 5 6 7 8 9

(d)

Wavelength (nm)

log(I)

Figure 2: Plaque count agar spectra: (a) raw UV-VIS; (b)

MSC UV-VIS; (c) raw VIS-NIR; (d) MSC VIS-NIR.

ance, the normalized media spectra matrix was sub-

tracted to the microorganisms spectra, obtaining the

spectra matrix (x), which is thereafter subjected to

robust mean scattering correction, and singular value

decomposition.

2.3.2 Robust Mean Scattering Correction

The collected spectra were smoothed by using a

Savisky-Golay ﬁlter (length = 4, Order= 2) (?) prior

to any exploratory data analysis procedure. After-

wards, the spectra was pre-processedusing a modiﬁed

multiplicative scatter correction algorithm (Gallagher

et al., 2005; Martens and Stark, 1991; Martens et al.,

2003). Each spectra is corrected by using the follow-

ing equation:

corr

= xb+ a = x

ref

(2)

The a and b are computed by minimizing the fol-

lowing error:

= bx

+ a− x

ref

(3)

where the x

is the j sample spectra and x

ref

is a

reference spectra.

This RMSC algorithm is based on the application

of the robust least squares method to determine the a

and b matrices ensuring that spectral areas that do not

correspond to scattering artifacts are not taken into

account. The robust least squares algorithm is im-

plemented by the re-weighted least squares with the

weights computed by using the Huber function. The

algorithm high breakdown point (50%) means that ex-

istent outliers will not distort the model ﬁtting (eq. 3)

and thus, the a and b scatter correction parameters are

determined using only consistent spectral areas. The

iterative algorithm can be described, brieﬂy as follow:

1) set the reference spectra (x

ref

) equal to the sample

−10 −5 0 5

−1.0 −0.5 0.0 0.5 1.0

PC1(96.40%)

PC2(2.03%)

(a)

TSA

−4 −2 0 2 4 6

−0.5 0.0 0.5

PC1(98.60%)

PC2(0.7%)

(b)

TSA

Figure 3: Growth media Gabriel plot: (a) UV-VIS (PC1

(96.40%), PC2 (2.3%) and (b) VIS-NIR (PC1 (98.60%),

PC2 (0.70%) to the corresponding growth media: TSA (),

MP (◦), LB (•) and YPD (⋄).

spectra closest to the median spectra; 2) correct the

remaining sample spectra by applying the above de-

scribed robust least squares procedure; eq.3) recom-

pute the median spectra and iterate until convergence.

2.3.3 Singular Value Decompostion

Singular value decomposition (SVD) is a blind signal

technique widely used in spectroscopy data, where

the corrected spectra (x

corr

) is decomposed in order of

magnitude of variation directions in the variable space

(wavelengths). Generally, most variability is captured

in the ﬁrst principal components (PC), where as, in

good signal to noise spectral data, noise is captured

in the last orthogonal decompositions, and therefore a

spectra can be decomposed as:

corr

x+ ε(x) (4)

where the

x is the signal and ε(x) is the estimated

noise of x. Spectra matrix x

corr

can be decomposed

by (SVD), where:

x = USV

(5)

where US are the scores, V

the loadings and

the S singular values, respectively (Jolliffe, 1986;

Krzanowski, 1998; Baig and Rehman, 2006).

To distinguish between the number of relevant de-

compositions, one can determine the number of rele-

vant singular values by performing n randomizations

of the original spectra matrix (x) (Manly, 1998). In

this research, 5000 randomizations were performed

by rotating the spectral scope value at the same wave-

lengths among the different samples, in order to not

violate the spectral continuity. Singular values from

the original spectra x above the 1st singular value of

randomized spectra (x

rand

) deﬁne the number of inde-

pendent singular values of the original signal where is

possible to discriminate the different microorganisms

spectra:

x = US

relv

(6)

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

400 500 600 700 800 900 1000

5 6 7 8 9

(a)

Wavelength (nm)

log(I)

400 500 600 700 800 900 1000

5 6 7 8 9

(b)

Wavelength (nm)

log(I)

400 600 800 1000

3 4 5 6 7 8 9

(c)

Wavelength (nm)

Log(I)

400 500 600 700 800 900 1000

3 4 5 6 7 8 9

(d)

Wavelength (nm)

log(I)

Figure 4: Microorganisms spectra: (a) raw UV-VIS; (b)

MSC UV-VIS; (c) raw VIS-NIR; (d) MSC VIS-NIR.

Where US

relv

and V

relv

are the statistically rele-

vant scores and loading of x, respectively. To further

discriminate between the microorganisms spectra, the

relevant PC’s scores(US

relv

) were subjected to hierar-

chical clustering according to the euclidean distance,

to determine the potential of using UV-VIS and VIS-

NIR to identify the studied yeast and bacteria. All

statistical computing analysis were performed using

R (R-Project, 2006).

3 RESULTS AND DISCUSSION

3.1 Spectral Analysis

Figure 2 presents the UV-VIS and VIS-NIR plate

count agar growth media spectra, respectively. It is

possible to observe in Figure 2(a) and 2(c) that all

plate count agar are highly dispersive, generating a

high scattering effect. This undesired scattering ef-

fect is due to the light path length to be very sensitive

to the probe angle, particles in the agar, surface tex-

ture of both agar and petri disk. If the light scattering

effect is not corrected, variance due to this physical

phenomenon affects signiﬁcantly the chemical inter-

pretation of the spectra due to scattering artifacts.

As pure additive effects of light scattering are

rarely observed in samples with complex compo-

sitions, being mainly of multiplicative origin, the

growth media spectra was subjected to RMSC, be-

ing the corrected spectra presented in Figures 2 (b)

and 2 (c), respectively. By directly comparing Fig-

ures 2 (a)-(b), and 2 (c)-(d), one can observe the scat-

tering effect is obtained in both light sources. After

applying the RMSC one can observe that, this scat-

tering artifacts are signiﬁcantly reduced in the region

of 700-1000nm, but nevertheless both light sources

present higher degree of spectral variance in the re-

gion of 400-700nm.

Furthermore, it is observable that all spectra are

proportional to each other. Variation is mostly in

terms of signal intensity than in spectra shape. In this

sense it is difﬁcult to distinguish the different growth

media by direct spectra comparison. Figure 3 presents

the Gabriel plot (PC1 vs PC2) of the growth media

spectra, for UV-VIS and VIS-NIR wavelengths, re-

spectively. Both UV-VIS and VIS-NIR biplots evi-

dence optical properties of the growth media, being

thin mostly described by the 1st principal component.

In both biplots, it is possible to observe that the MP

growth media is the most translucid and YPD the

most opaque, and therefore PC1 can be interpreted as

the amount of signal that the detector records. Media

such as TSA and LB present similar spectra records.

Such is mainly attributed to their similar composition

in terms of main componentssuch as sodium chloride,

agar and water.

Results show that MP and YPD media are more

suceptible to variability than TSA and LB in spec-

troscopy terms. As no chemical assessment was per-

formed to the media, we cannot present the cause for

this source of variation.

Figure 4 presents the UV-VIS and VIS-NIR mi-

croorganisms spectra, respectively. Similarly to the

growth media, the microbe spectra exhibits high scat-

tering artifacts (see Figures 4a and 4c). The scatter-

ing effect is in this case due to the light scattering at

the colony surface and growth media, which signiﬁ-

cantly affects the observed spectra. Similarly to the

growth media, scattering is signiﬁcantly high in both

light sources and evenly distributed from 450 to 1000

nm, nevetheless scattering is signiﬁcantly reduced by

the RMSC algorithm. The corrected spectra presents

higher variability in the region of 400-700nm, but

nevertheless it is difﬁcult to recognize directly spec-

tral characteristics that distinguish the different mi-

croorganisms under study, and therefore SVD analy-

sis is necessary.

3.2 Singular Value Decomposition

Analysis

Figure 5 presents the Gabriel plot of the ﬁrst

two components obtained by SVD (PC1(78.40%),

PC2(8.03%)), and the corresponding hierarchical

clustering analysis. The ﬁrst two decompositions en-

sure the majority of the spectral variation to pro-

portional to the average spectrum (PC1(78.40%)),

and that linear variance also with discriminant power

(8.03%), evidences smaller but discriminate spectral

FEASABILITY OF YEAST AND BACTERIA IDENTIFICATION USING UV-VIS-SWNIR DIFUSIVE

REFLECTANCE SPECTROSCOPY

−10 −8 −6 −4 −2 0 2 4

−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5

PC1(78.40%)

PC2(8.03%)

(a)

M.Luteus

P.Flurescencis

E.Coli

B.Cereus

S.Cerevisiae

S.Bayanus

C.Albicans

Y.Lipolytica

M.luteus

E.coli

C.albicans

B.cereus

S.cerevisiae

C.albicans

S.cerevisiae

C.albicans

S.cerevisiae

B.cereus

C.albicans

S.cerevisiae

Y.lipolytica

S.cerevisiae

C.albicans

S.cerevisiae

C.albicans

B.cereus

C.albicans

S.cerevisiae

C.albicans

S.cerevisiae

C.albicans

S.cerevisiae

C.albicans

B.cereus

C.albicans

S.cerevisiae

C.albicans

B.cereus

E.coli

S.bayanus

Y.lipolytica

S.bayanus

Y.lipolytica

E.coli

S.bayanus

Y.lipolytica

S.bayanus

Y.lipolytica

S.bayanus

Y.lipolytica

0 5 10

(b)

Height

Figure 5: UV-VIS Microorganisms Spectra PCA Analysis: (a) Gabriel Plot (PC1 (78.40%), PC2 (8.03%); S. cerevisiae (⋄), S.

bayanus (△), C. albicans (▽), Y. lipolytica (H), M. luteus (•), P. ﬂuorescens (◦), E. coli () and B. cereus (); (b) hierarquical

clustering of Microorganisms.

differences between the studied microorganisms.

PC1(78.40%) discriminates between spectral in-

tensity, being possible to observe that colonies of

B. cereus, C. albicans and S.cerevisiae, present the

higher scores indicating that the colonies of this mi-

croorganisms are well suited for diffusive reﬂectance.

Although the relative proximity in the scores space, it

is observable that B. cereus, C. albicans and S. cere-

visiae are discriminated by the 2nd PC. In the group,

it is possible to observe a higher similarity between C.

albicans and S.cerevisiae spectra than with B. Cereus.

It is further observable that although S. bayanus

and Y. lipolityca exhibit similar spectral intensity,

their spectra is possible to he discriminated in the 2nd

PC. S. bayanus who presents larger variability then Y.

lipolityca, being more difﬁcult to identify.

E. coli and M. luteus colonies are distinguish-

able from all the other microorganisms group. These

present the lowest signal intensity, and the proximity

of the two spectra may in part he due to the growth

media spectral similarity between TSA and LB, as

show in Figure 3a. Nevertheless, the UV-VIS spectra

decomposition is capable of discriminating between

E. coli and M. luteus spectra.

In the UV-VIS, P. ﬂuorescens spectra has proved

to he highly unreproducible, being its scores

well spread throughout the 2nd PC. This un-

reproducibility is attributed to the high translucent

of both P. ﬂuorescens colonies and the MP growth

media, and to the experimental microbiological tech-

nique. As microorganisms were inoculated using a

inoculating loop, a signiﬁcant part of radiation is dif-

fused into de media in P. ﬂuorescens. Better growth

may in the future improve the spectral measurement

and therefore its identiﬁcation, as already observed in

E. coli colonies, which although are smaller, are ca-

pable of developing thicker colonies.

Dispite the experimental difﬁculties hierarchical

clustering analysis presented in Figure 5b shows that

the majority of the studied microorganisms cluster to-

gether with exception of P. ﬂuorescens. This gives

good perspectives of using UV-VIS spectroscopy for

microorganisms identiﬁcation in plate count agar af-

ter experimental and signal processing improvements.

The VIS-SWNIR spectra also exhibits high scat-

tering artifacts in the 400-1000nm region (see Fig-

ure 4c), the scattering effect is effectively removed

by the RMSC algorithm used during the spectral pre-

processing. Figure 4d, shows that in the corrected

spectra, variance is higher in the 400-700nm as al-

ready observed as with the UV-VIS light source. The

majority of the spectra are proportional to each other,

varying only in signal intensity.

Figure 6a presents the Gabriel plot of the ﬁrst

two PCs of the decompose VIS-NIR spectra (PC1

(75.40%), PC2 (10.95%), and the corresponding

scores hierarchical clustering. Similarly to the UV-

VIS spectra, the ﬁrst spectra decomposition discrim-

inating variance proportional to the average spec-

tra, and PC2 (lower variance) tends to discriminate

smaller details between the spectra of the studied mi-

croorganisms, evidencing that VIS-NIR is a viable

methodology for identiﬁcation microorganisms on

plate count agar. PC1 (75.40%) of VIS-SWNIR spec-

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

−15 −10 −5 0 5

−3 −2 −1 0 1 2

PC1(75.40%)

PC2(10.95%)

(a)

M.Luteus

P.Flurescencis

E.Coli

B.Cereus

S.Cerevisiae

S.Bayanus

C.Albicans

Y.Lipolytica

E.coli

M.luteus

P.fluorescens

M.luteus

P.fluorescens

M.luteus

P.fluorescens

E.coli

B.cereus

S.bayanus

S.cerevisiae

S.bayanus

S.cerevisiae

C.albicans

B.cereus

C.albicans

Y.lipolytica

B.cereus

C.albicans

B.cereus

Y.lipolytica

B.cereus

C.albicans

0 5 10 15

(b)

Height

Figure 6: VIS-NIR Microorganisms SpectraPCA Analysis: (a) Gabriel Plot (PC1 (75.40%), PC2 (10.95%); S. cerevisiae.

(⋄), S. bayanus (△), C. albicans (▽), Y. lipolytica (H), M. luteus (•), P. ﬂuorescens (◦), E. coli () and B. cereus (); (b)

hierarquical clustering of Microorganisms.

tra presents a clear discrimination between groups

of microorganisms: (a) B. cereus, S.cerevisiae, S.

bayanus, Y. lipolityca, C. albicans, and (b) M. luteus,

E. coli and P. ﬂuorescens. PC2 (10.95%) is capa-

ble of discriminate the microorganisms inside these

two groups, and therefore, complete discrimination is

achievable with VIS-SWNIR wavelengths.

In the VIS-SWNIR spectra E. coli and P. ﬂuo-

rescens present higher dispersion of signal when com-

pared to the rest of the spectra of microorganisms.

Nevertheless, results reproductivity inside these two

groups is signiﬁcantly higher than with the UV-VIS

measurements; indicating that translucid colonies

were better identiﬁed under the VIS-SWNIR radi-

ation. Furthermore, data suggest that VIS-SWNIR

may be better to be used under non-optimal measur-

ing conditions. This is especially problematic if we

want to identify microorganisms in plate count agar

with similar compositions and metabolism.

3.3 Methodology Improvements

This preliminary study shows that UV-VIS and VIS-

SWNIR reﬂection spectroscopy has a great potential

for rapid qualitative discrimination of yeasts and bac-

teria in plate count agar. Nevertheless, both experi-

mental methodologyand signal processing techniques

should be improved to take advantage of the informa-

tion contained in the UV-VIS-SWNIR.

Improvements to the experimental methodology are

necessary to improve discrimination in the studied re-

gion of the spectra. For example, the use of liquid

cultures to replicate microorganisms and then innoc-

ulate by a droplet on the surface agar will allow the

growth of compact and thicker colonies. Small and

thin colonies lead to readings with much media infor-

mation, because the light is reﬂected from both sites

of the agar. This is particularly relevant for E. coli

and P. ﬂuorescens because of the small size of the

colonies and small thickness of P. ﬂuorescens; being

the spectra highly affected by the media composition,

leading to sistematic errors in discrimination. Mi-

croorganisms signal spectra can also be maximised

by reducing the growth media thickness. Such min-

imises light dispersion accross the agar, and increases

the colony spectral intensity by passing through it the

reﬂected light; ans as well by optimising ﬁber optics

diffusive reﬂectance position and angle control, min-

imizing scattering effects in the colonies and growth

media.

Scattering artifacts and small noise were success-

fully removed by pre-processing the spectra with the

RMSC and Savisky-Golay ﬁlter, respectively; being

possible to achieve high-quality and resolution ﬁnal

spectra before signal treatment. Improvements can

also be performed to the spectra processing proce-

dure. Methods such as the combination of logis-

tic partial least squares (log-PLS) with multiblock

(e.g. UV-Vis+Vis-Nir spectra) can provide a new in-

sight to the discrimination of microorganisms using

spectroscopy. If these methodologies provide good

discriminations, more robust techniques such as the

use of wavelets for compressing the original spectra

and orthogonal-PLS classiﬁcation of spectra will be

tested.

FEASABILITY OF YEAST AND BACTERIA IDENTIFICATION USING UV-VIS-SWNIR DIFUSIVE

REFLECTANCE SPECTROSCOPY

4 CONCLUSIONS

Results show that UV-VIS-SWNIR spectroscopy is a

feasable technology for plate count agar microorgan-

isms identiﬁcations. The robust mean scattering cor-

rection algorithm was able to efﬁciently remove the

growth media and colonies scattering artifacts, allow-

ing a better interpretation of the singular value de-

composition scores loading. In this exploratory ex-

periment, VIS-SWNIR wavelengths were able to pro-

duce better discriminations between microorganisms

than the UV-VIS region. Nevertheless, experimen-

tal methodology and signal processing improvements

proposed may increase the discrimination resolution,

making UV-VIS-SWNIR an attractive methodology

for rapid microorganisms identiﬁcation in plate count

agar.

REFERENCES

Avantes (2007). Avaspec-2048-4-dt/-rm.

http://www.avantes.com.

Baig, S. and Rehman, F. (2006). Signal modeling using sin-

gular value decomposition. In Advances in Computer,

Information, and Systems Sciences, and Engineering.

Springer Netherlands.

Berger, A. and Zhu, Q. (2003). Identiﬁcation of oral bacte-

ria by raman microspectroscopy. In Journal of Mod-

ern Optics 50(15-17): 2375-2380. Taylor and Francis

Group.

Bhatta, H., Goldys, E., and Learmonth, R. (2005). Rapid

identiﬁcation of microorganisms by intrinsicﬂuores-

cence. In Imaging, Manipulation, and Analysis of

Biomolecules and Cells: Fundamentals and Applica-

tions III,. SPIE.

Difco, L. (2005). Difco manual : dehydrated culture media

and reagents for microbiology. - VII. Difco Laborato-

ries, Detroit, USA.

Dziuba, B., Babuchowski, A., Naleczb, D., and Niklewicz,

M. (2007). Identiﬁcation of lactic acid bacteria using

ftir spectroscopy and cluster analysis. In International

Dairy Journal 17: 183189. Elsevier.

Gallagher, N. B., Blake, T., and Gassman, P. (2005). Ap-

plication of extended inverse scattering correction to

mid-infrared reﬂectance of soil. In Journal of Chemo-

metrics 19: 271-281.

Guibeta, F., Marieya, L., Pichonb, P., Traverta, J., Denisb,

C., and Amiela, C. Discrimination and classiﬁca-

tion of enterococci by fourier transform infrared spec-

troscopy.

Hammes, G. G. (2005). Fundamentals of spectroscopy. In

Spectroscopy for the Biological Sciences. John Wiley

Sons, Inc.

Harz, M., Rosch, P., Peschke, K.-D., Ronneberger, O.,

Burkhardtb, H., and Popp, J. (2005). Micro-raman

spectroscopic identiﬁcation of bacterial cells of the

genus staphylococcus and dependence on their cul-

tivation conditions. In Analyst 130: 15431550. The

Royal Society of Chemistry.

Jolliffe, I. (1986). Principal Component Analysis. Springer,

New York, USA.

Krzanowski, W. J. (1998). Principles of Multivariate Data

Analysis. Oxford University Press, Oxford, UK.

Manly, B. F. (1998). Randomization, Bootstrap and Monte

Carlo Methods in Biology. Chapman and Hall, Lon-

don, UK, 2nd edition.

Mariey, L., Signolle, J., Amiel, C., and Travert, J. (2001).

Discrimination, classiﬁcation, identiﬁcation of mi-

croorganisms using ftir spectroscopy and chemomet-

rics. In Vibrational Spectroscopy 26: 151-159. Else-

vier.

Martens, H., Nielsen, J. P., and Engelsen, S. B. (2003).

Light scattering and light absorbance separated by ex-

tended multiplicative signal correction. application to

near-infrared transmission analysis of powder mix-

tures. In Analytical Chemistry 75(9): 394-404. Amer-

ican Chemical Society.

Martens, H. and Stark, E. (1991). Extended multiplicative

signal correction and spectral interference subtraction:

new preprocessing methods for near infrared spec-

troscopy. In Journal of Pharmaceutical and Biomedi-

cal Analysis 9: 625-635. American Chemical Society.

Oberreuter, H., Mertens, F., Seiler, H., and Scherer, S.

(2000). Quantiﬁcation of micro-organisms in binary

mixed populations by fourier transform infrared (ft-

ir) spectroscopy. In Letters in Applied Microbiology

2000, 30: 8589. The Society for Applied Microbiol-

ogy.

Oberreuter, H., Seiler, H., and Scherer, S. (2002). Iden-

tiﬁcation of coryneform bacteria and related taxa by

fourier-transform infrared (ft-ir) spectroscopy. In In-

ternational Journal of Systematic and Evolutionary

Microbiology 52: 91100. IUMS.

Perkauparus, H., Grinter, H., and Therfall, T. (1994). Uv-Vis

spectroscopy and its applications. Springer-Verlag,

New York, USA.

R-Project (2006). R: A programming environment for data

analysis and graphics. URL: http://www.r-project.org.

Rch, P., Harz, M., Schmitt, M., and Popp, J. (2005). Raman

spectroscopic identiﬁcation of single yeast cells. In J.

Raman Spectrosc 36: 377379. Wiley InterScience.

Riddle, J., Kabler, P., Kenner, B., Bordner, R., Rockwood,

S., and Stevenson, H. (1956). Bacterial Identiﬁcation

by Infrared Spectrophotometry. Cincinnati, Ohio.

Winder, C., Carr, E., Goodacre, R., and Seviour, R. (2004).

The rapid identiﬁcation of acinetobacter species using

fourier transform infrared spectroscopy. In Journal

of Applied Microbiology 96, 328339. The Society for

Applied Microbiology.

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing