Comoescreverumartigocientífico

(Parte3)

AntônioC.Roque

Tabelasefiguras

•  Oobjetivodastabelasefigurasémostrardadosquesejammuitonumerososoucomplicadosparaseremdescritosadequadamentenotextoe/ourevelartendênciasoupadrõesnosdados.

•  Astabelasefigurassãocríticas:seoleitorforalémdoAbstract,éprovávelqueeleexamineastabelasefigurasemseguida.

Estudecomoapresentarseusdadosresultados

•  Nemtodososdadoseresultadosprecisamsermostradosemformadetabelasougráficos.

•  Algunsficambemnotexto,sumarizadosentreparênteses:Seedproductionwashigherforplantsinthefull-suntreatment(52.3±6.8seeds)thanforthosereceivingfilteredlight(14.7±3.2seeds,t=11.8,df=55,p<0.001).

Exemplotiradode:http://abacus.bates.edu/~ganderso/biology/resources/writing/HTWtoc.html

Comofazerboastabelasefiguras

•  Decidaqueresultadosficammelhormostradoscomotabelasefigurasdoquecomotexto.

•  Limiteonúmerodetabelasefigurasàquelasqueforneceminformaçãoessencial.

•  Incluaapenasresultadosrelevantesparaasquestõescolocadasnaintrodução,independentementedeapoiaremounãoashipóteses.

•  Desenhecadatabelaefiguraparaquesejaminteligíveisporsisó,semreferênciaaotexto.

•  Enumerecadafiguraetabelanaordememqueelassãoreferidasnotexto(asfiguraseastabelastêmnumeraçõesseparadas).

•  Organizeastabelasefigurasemumaordemtalqueelascontemumahistória.

•  Verifiquenasinstruçõesparaautoresdarevistaondecolocarastabelasefiguras.

•  Dependendodarevista,elasdevemvirempáginasseparadasnofimdoartigo(depoisdasreferências)ounoslocaisapropriadosaolongodotexto.

•  Casoastabelasefigurassejamincorporadasaotexto,certifique-sedequenãoháumaquebradepáginanomeiodeumadelas.

•  Façaumtítuloparacadatabelaeumalegendaparacadafigura.

•  Dependendodarevista,ostítuloselegendasdevemserlistadosseparadamentenofimoucolocadosacimadatabelaeabaixodafigura,respectivamente.

•  Ostítulosdastabelaseaslegendasdasfigurasdevemfornecerinformaçãosobreoqueémostradonelas,masnãoumresumoouumainterpretaçãodosresultados.

•  Certifique-sedequetodasasfigurasetabelasestejamreferenciadasnotexto.

•  Casosejamincluídasfigurasetabelaspreviamentepublicadas,obtenhaaspermissõesdosproprietáriosdosseuscopyrights(emgeraloseditores)ecoloqueumagradecimentoaeles.

Comosereferiratabelasefigurasnotexto•  Usefrasesquechamemaatençãodoleitorparaarelaçãoou

tendênciaquevocêquerdestacar,comareferênciaàtabelaoufiguraentreparênteses:

Germinationratesweresignificantlyhigherafter24hinrunningwaterthanincontrols(Fig.4).DNAsequencehomologiesforthepurplegenefromthefourcongeners(Table1)showhighsimilarity,differingbyatmost4basepairs.

•  Evitefrasesqueapenasindiquemaoleitorqueosdadosouresultadosobtidosestãomostradosemumatabelaoufigura:

Table1showsthesummaryresultsformaleandfemaleheightsinthesample.

Tabelas

•  Tabelassãousadasparatornarumartigomaisagradáveldeselerporretirardadosnuméricosdotexto.

•  Tabelastambémsãousadasparasintetizaraliteraturaexistente,paraexplicarasvariáveisouparaapresentarostermosusadosemenquetesfeitas.

•  Crietabelascomaferramentadetabeladoeditordetexto.Nãomontetabelasamão.

•  Façacabeçalhos(rótulos)paraascolunasdastabelasdemaneiraqueoconteúdodecadacolunasejaevidentesemnecessidadedeseconsultarotexto.

•  Verifiquenasinstruçõesparaautores,masamaioriadasrevistaspedequecadatabelaestejajuntocomoseutítuloemumapáginaseparada.

Figuras

•  Asfigurastêmimpactovisuale,portanto,muitasvezessãoamelhormaneiradesecomunicaroresultadoprincipal.

•  Asfigurassãotradicionalmenteusadasparamostrartendênciaseresultadosdegrupos,maselastambémpodemserusadasparailustrarprocessosoumostrardadosdetalhadosdeformasimples.

•  Rotulecadaeixoincluindoasunidadeseidentifiqueclaramenteosdadosmostrados(porexemplo,rotulecadalinhamostradaemumgráfico).

•  Verifiquenasinstruçõesparaautores,masamaioriadasrevistaspedequeaslegendasdasfigurassejamlistadasemordemnuméricaemumapáginaseparadaequecadafiguraestejasozinhaemumapáginaseparada.

•  Asfigurasdevemserdealtaqualidade.Verifiquenasinstruçõesparaautoresqueformatodeimagemépreferidopelarevista.

•  Emgeral,asfigurassãoempretoebranco.Ousodecorémuitocaroparaoseditoresecostumasercobradodosautores.

•  Figurascoloridassódevemserusadasquandoessenciais.

•  Nãoincluadetalhesexperimentaisnaslegendasdasfiguras.Essesdetalhesdevemficarnaseçãodemétodos.

•  Sóincluafotosdepessoascasotenhaobtidoautorizaçãoporescritodelas.

•  Escolhaaformacorretaparacadafigura:– Seasvariáveisindependenteedependenteforemnuméricas,usediagramasdedispersão;

– seapenasavariáveldependentefornumérica,useumgráficodebarras(inclualinhashorizontaisparaindicarmédias,desviospadrõeseintervalosinterquartis);

– paraproporções,usegráficosdebarrasoudetortaoupizza.

•  Procureexemplosdessestiposdefigurasnosartigosquevocêlê.

ExemplosdeTabelaseFiguras

MeasuresTo quantify observed spatiotemporal patterning of network activity anddistinguish between various network behaviors, we applied three mea-sures: average frequency ( F), mean phase coherence ( R), and a measureof synchronous bursting ( B). The combination of these three measuresallowed us to compare network dynamics quantitatively, and also detectbehavior switching within a single simulation run.

Frequency. The average frequency of cell n, Fn, was defined as theinverse of the average interspike interval over the duration of the simu-lation run:

Fn !1

"n, "n ! !

k!1

S"1tk#1 # tk

S # 1,

where S is the total number of spikes fired at times tk of cell n. Thenetwork average frequency, F, was the average of Fn over the number ofcells in the network.

Mean phase coherence. To quantify phase locking between cells, weadapted the mean phase coherence, R, of an angular distribution (Mor-mann et al., 2000). The value of R ranged between 0 and 1, and increasedas phase locking increased between cells. We measured the time depen-dence of R with a sliding window of 750 ms. R is defined as follows:

R ! "1

S!j!0

S"1

ei$nm" ,

where S denotes the number of samples in the array of cell n spike times,and $n,m is the phase between cells n and m for interspike interval j. Thiswas determined as follows: the period for interspike interval j,"nj # tnj#1 # tnj, for cell n was taken to be 2%. The cell m spike associatedwith interval j, tmj, was selected such that tnj & tmj#1 & tnj#1 so the phasebetween spikes at time tnj and tmj (interval "nj,mj # tmj # tnj) could becalculated at time tnj by $n,m # $"nj,mj/"nj%2%.

Synchronous bursting. We used an interspike distance synchrony mea-sure (Tiesinga and Sejnowski, 2004) to monitor the degree of spikingsynchrony in the network. The metric, B, is based on the time-ordered,complete set of network spikes and relies on the fact that the variancebetween firing times of all cells in the network during a synchronousevent is smaller than during asynchronous events. B is defined as follows:

B ! $ % &"'2' # &"''2

&"''# 1& 1

% N,

where N is the number of cells in the network. The combined, time-ordered set of network spike times t' was labeled by the index (, whereasthe set of network interspike intervals was labeled "' with "' ! t' # 1 " t'.Note that these interspike intervals are between different cells in thenetwork. Thus, assuming that every neuron fires independently with aconstant rate, the combined spike train for a large asynchronous networkwill have a Poisson spike distribution with the term% &"v

2' # &"v'2/&"v'3 1. However, in the limit of large N and if the net-work is fully synchronized with neurons firing with a period T, the term% &"v

2' # &"v'2/&"v'3 % N. Thus, the relatively atypical form of B pro-vides a normalized measure of degree of synchronized bursting in thenetwork, where low values of B are indicative of asynchronous activity,whereas B ! 1 indicates strong, highly synchronous bursting.

Parametric distance. To determine overall dissimilarity between net-

work states, we formulated a parametric distance, D, between simulationruns 1 and 2 as determined from all three measures:

D2,1 ! % $B2 # B1

B2 ) B1& 2

) $R2 # R1

R2 ) R1& 2

) $F2 # F1

F2 ) F1& 2

.

D was small if the behavior of runs 1 and 2 was similar and was largebetween dissimilar runs.

ResultsModel cell excitability propertiesBy modulating the activation characteristics of the delayed recti-fier K# current and the maximal conductances of the ionic cur-rents, we created four model cells having various membrane ex-citability properties as described by their f–I curves (Fig. 1). Bothcells A and B had characteristic type I membrane excitabilityproperties (Rinzel and Ermentrout, 1998; Izhikevich, 2001).They displayed a continuous f–I curve indicating the appearanceof arbitrarily low firing frequencies at firing threshold. Cells Cand D exhibited type II excitability with a nonzero, “critical”firing frequency at threshold (Rinzel and Ermentrout, 1998). An-other distinguishing characteristic of type I and II excitability isthe slope of the f–I curve at high applied current: uniformity offiring frequency at high input currents is typical of type II cells.Thus, cell B, whose f–I curve at high current tended toward ashallower slope, was less type I-like than cell A, which had asteeper f–I slope typical of type I excitability. However, cell C hada lower critical firing frequency at threshold than cell D and dis-played a steeper f–I slope, reminiscent of type I excitability. Thus,with these four cells types, we were able to explore network effectsresulting from a transition in neuronal excitability from type I totype II behavior. Model parameter values that were varied tocreate the four cell types are listed in Table 1.

Phase response curve analysisFor periodically firing neurons, phase response curves (PRCs)describe how small, brief inputs given at different phases of theperiodic cycle affect the timing of subsequent spikes (see supple-mental material, available at www.jneurosci.org). It has been sug-gested that PRCs may help to elucidate the mechanisms by whichsome cells tend to synchronize when coupled, whereas otherstend toward antisynchrony (Hansel et al., 1995; Ermentrout,1996; Izhikevich, 1999; Ermentrout et al., 2001). To obtain thephase response curves for our model cells, we elicited a fixedbackground firing frequency by injecting an appropriate appliedcurrent (as determined by the f–I curve of the cell). We theninjected small, EPSP-like inputs at different times between peri-odically occurring spikes. Figure 2 shows the PRCs for all cellmodels with background firing at 40 Hz. The EPSP-like stimuluswas a current pulse with amplitude of 0.21 nA/cm 2 and durationof 2 ms.

There was an obvious shift in the shape of the PRC as cellstransitioned from type I-like to type II-like. Type I-like cells A andB had positive PRCs and displayed an advance in spike firing

Table 1. Model cell parameters altered to construct four cell types

Vhalf K#dr (mV) gKdr (S/cm2) gNa (S/cm2) gKa (S/cm2) gh (mS/cm2) EK (mV) ENa (mV) Eh (mV)

Cell A 13 0.2 0.3 0.048 0.5 "90 55 "30Cell B 13 0.5 0.5 0.048 0.1 "77 50 0Cell C 0 0.08 0.3 0.048 0.75 "90 55 "30Cell D 0 0.7 1.5 0.03 0.5 "77 50 0

Changes in the steady-state, half-activation of the Kdr current (Vhalf K#dr) eliminated low-frequency firing at threshold for type II-like cells. Changes in maximal conductances of Na# (gNa) and K#-dr (gKdr) currents promoted spiking.

Changes in maximal conductances of K# A-type (gKa), and h (gh) currents modulated resting membrane potential and rheobase current.

Bogaard et al. • Synergy of Cellular and Network Mechanisms J. Neurosci., February 11, 2009 • 29(6):1677–1687 • 1679

Bogaard et al. (2009), J. Neurosci. 29:1677-1687

Titulo da tabela (Table legend)

Rótulos das colunas

Corpo da tabela (dados)

Linha demarcatória (separa as diferentes partes da tabela) Notas à tabela (Table footnotes)

Isto não faz parte da tabela (é só a referência de onde ela foi tirada)

found only at one side of the central peak, then the CCG was classifiedas group 2. If the negative trough was found at neither side of thecentral peak, then the CCG was classified as group 3. The parametersof the algorithm were chosen so that the automated classificationmatched well with the classification made by visual examination. AnACG was defined to have negative flanks if it had a negative troughwithin the time lags from 5 ms to 25 ms, which undershot by !5 SDsthe mean in the baseline intervals at lags from 25 ms to 125 ms. Wedenoted the time lag of the minimum as Tmin. We admitted ACGnegative flanks only if the value at Tmin was more negative than itstwo nearest neighbors at the side toward time lag zero so that therewas a local maximum between Tmin and zero time lag.

For each pair of neurons, we computed the time course of spike-count noise correlation rsc within a 100-ms time window sliding at a10-ms step. Before computing rsc, we converted the data into z-scoresto normalize spike counts. To avoid some of the possible artifacts ofcorrelation analysis, we removed trials on which the response of eitherneuron of a pair was !5! different from its mean response, and wecomputed rsc only if each neuron of a pair yielded at least 12 distinctvalues of spike count.

For each MT neuron, we also computed the time course of the Fanofactor in 100-ms intervals as the variance of spike count divided bythe mean spike count. For neurons that were tested with random-dotstimuli moving in multiple directions and speeds, the preferred direc-tion (PD) and the PS of the neurons were determined using the samemethods as we described previously (Huang and Lisberger 2009). Thedifference of the PDs and the log ratio of the PSs were used tocharacterize the separations between two neurons’ stimulus preferences.

Computational model. We constructed a simple neural networkmodel using excitatory and inhibitory integrate-and-fire neurons. In agiven experimental trial, the activity of each input neuron was mod-eled as a spike train whose interspike intervals followed a Poissondistribution with a given mean. The membrane potential Vp (p " 1,2)of the excitatory model neurons was determined by

Cm

dVp

dt" # !

j"1

Nex

Gex"Vp(t) # ENa# # !j"1

Nin

Gin"Vp(t) # ECl#

# !j"1

2

Gie"Vp(t) # ECl# # Gleak"Vp(t) # Vrest## GAHP"Vp(t) # EAHP# # Gadp"Vp(t) # Eadp#

(3)

where Gex is the feedforward (FF) excitatory conductance, Gin is theFF inhibitory conductance, and Gie is the intracortical inhibitoryconductance. The membrane potential of the inhibitory model neuronsused the same equation without the last term [adaptation (adp)

conductance Gadp] and with Gie replaced by Gei, the intracorticalexcitatory conductance between the excitatory and inhibitory neurons.We modeled each conductance as

Gx " Gx_max!p

e# (t# tp# td)

$x (4)

where tp is the timing of a spike p in the presynaptic neuron for eachinput conductance, in the postsynaptic neuron for Gleak (leaky con-ductance), GAHP (afterhyperpolarization conductance), and Gadp.Other parameters are: td, a synaptic delay of 2.5 ms for Gex, Gin, Gie,Gei and a time delay of 1 ms for Gleak, GAHP, and Gadp; Gx_max is themaximum conductance. The default values of the model parametersare listed in Table 1 [also see Shadlen and Newsome (1998); Somerset al. (1995, 1998)]. Parameter values that deviated from the defaultvalues are mentioned in RESULTS.

Spikes occurred in a model neuron when its membrane potentialexceeded its spiking threshold Vthresh, with the caveat that 1) an actionpotential could not occur within the absolute refractory period, and2) Vthresh increased after the absolute refractory period and thendecayed exponentially (Eq. 5) to create a relative refractory period[after Somers et al. (1998)]

Vthresh(t) " Vthresh(t0) % Vthresh_elv e# (t# tspk)

$thresh_elv (5)

Because fast-spiking neurons show little spike-frequency adapta-tion (McCormick et al. 1985), we chose not to apply thresholdelevation to the inhibitory cortical neurons and relied on an outwardAHP current (IAHP) to model their modest relative refractory period.

In simulations that examined the effects of intrinsic outwardcurrents on spike-timing correlations, we removed connections fromthe inhibitory to the excitatory model neurons and added a potassium(K#) current

Ik " Gk"Vp(t) # EK# (6)

Here, Gk is determined by Eq. 4 and represents either the calcium(Ca2#)-activated K# current of small (SK) conductance (Stocker2004) or a hypothetical K# conductance Gh. Gh has artificiallydetermined magnitude and kinetics that were not constrained byexperimental data. The goal of introducing Gh was to simulate anAHP that mimicked the time course and amplitude of the inhibitorypostsynaptic potential (ipsp) generated by intracortical inhibition.

We conducted model simulations using Matlab (MathWorks,Natick, MA). Numerical solutions of the differential equations wereobtained using the fourth-order Runge-Kutta method at a time step of

Table 1. Default model parameters

Variable E I Variable E I Variable E and I

Cm 0.33 nF 0.17 nF Gff_e_max 6 nS 6 nS Vrest $70 mvGei_max 6 nS Gff_i_max 11.25 nS 11.25 nS ENa 55 mvGie_max 0 % 30 nS $ 2 ms 4 ms ECl $75 mvGAHP_max 40 nS 20 nS Eadp $90 mv EAHP $90 mvGleak_max 25 nS 20 nS $adp 30 ms EK $90 mvGadp_max 3 nS $spk 0.5 ms 0.25 ms $AHP 1 msNff_ex 100 30 Vthresh_base $55 mv $60 mv APamp 55 mvNff_in 100 30 Vthresh_elv 10 mv (sum & 20 mv) Abref 2 msCommonNff_ex 40% 0% $thresh_elv 8 ms td 2.5 or 1 msCommonNff_in 40% 0% FF input rate 2 % 30 Hz 1 Hz

E, excitatory neuron; I, inhibitory neuron; Cm, membrane capacitance. Gx_max, maximum conductance, where Gx can be Gei, conductance of intracorticalexcitation [at the synapse from excitatory (e) neuron to inhibitory (i) neuron]; Gie, conductance of intracortical inhibition; GAHP, afterhyperpolarization (AHP)conductance; Gleak, leaky conductance; Gadp, adaptation (adp) conductance. Nff_ex and Nff_in, number of excitatory and inhibitory feedforward (FF) input; CommonNff_ex and Nff_in, percentage of excitatory and inhibitory FF input that is shared by two neurons; Gff_e and Gff_i, conductance of the excitatory and inhibitory FFinput; $, membrane time constant; $adp and $AHP, time constants of adp and AHP conductance; td, synaptic time delay; APamp and $spk, amplitude and decayingtime constant of action potential; Abref, absolute refractory period; E, reversal potential of sodium (Na), chlorine (Cl), potassium (K), AHP, and adp current; Vrest,resting membrane potential; Vthresh_base, membrane potential of baseline spiking threshold; Vthresh_elv, elevation of spiking threshold; $thresh_elv, decaying timeconstant of elevated spiking threshold; FF input rate, firing rates of the input neurons.

853CIRCUIT MECHANISMS OF SPIKE-TIMING CORRELATIONS

J Neurophysiol • doi:10.1152/jn.00775.2012 • www.jn.org

at CAPES - U

sage on April 19, 2013http://jn.physiology.org/

Dow

nloaded from

Huang and Lisberger (2013), J. Neurophysiol. 109:851-866

051706-9 Bandos, Rockette, and Gur: Subject-centered free-response ROC (FROC) analysis 051706-9

TABLE II. Estimated power for a comparison of two diagnostic modalitiesunder the paired design using the permutation test. [The estimates are basedon 250 simulated dataset each of which is analyzed based on 250 randompermutations. The number of subjects in the groups with 0, 1, and 2 targets are120, 80, and 40, respectively. FROC parameters in the headings identify thesimulation scenario for the second diagnostic modality. The distribution ofthe output of the first diagnostic modality corresponds to TPF = 0.7 and P(X< Y) = 0.7, hence the corresponding cell provides the estimates for the type Ierror rate. For all scenarios the FPR is 2.0, and all characteristics had the samevalue for both groups of subjects with 1 and 2 targets. Considered values ofthe guessing parameter ϕ correspond to the probabilities 0.1, 0.3, and 0.6 ofhitting a target with a single mark in a sample with a given composition.]

P(X < Y)

Guessing parameter TPF Method 0.7 0.9

ϕ = 0.11 0.7 Target-centered 0.05 0.10Subject-centered 0.06 0.09

0.8 Target-centered 0.56 0.76Subject-centered 0.48 0.70

ϕ = 0.38 0.7 Target-centered 0.04 0.25Subject-centered 0.05 0.24

0.8 Target-centered 0.52 0.93Subject-centered 0.48 0.90

ϕ = 1.00 0.7 Target-centered 0.04 0.48Subject-centered 0.04 0.48

0.8 Target-centered 0.45 0.98Subject-centered 0.41 0.97

research.15, 16, 32 The general problem is frequently describedas a nonignorable cluster size, which cannot be adequatelyaddressed by the conventional marginal models.15 One ofthe solutions to this problem is to consider subject-specificrather than population-averaged estimates,15 which is asymp-totically equivalent to using marginal estimates weighted ac-cording to the cluster size.16 In this general context our ap-proach can be viewed as a simple model-free subject-centered(i.e., subject-specific) method for analysis of FROC data(which is frequently associated with nonignorable cluster-size), which can be implemented using reweighted target-centered (population-averaged) FROC estimates.

An alternative to the conventional FROC method of perfor-mance analyses in detection and localization task is providedby the region-of-interest (ROI) approach.30, 31 However, dueto the use of marginal probabilities in defining rates for clus-tered ROC data, the summary indices of the ROI approachare also target-centered. The subject-centered reformulationof the ROI performance curve and related indices could bedeveloped using the approach similar to the one applied inthis paper.

The developed overall summary index of the subject-specific FROC curve can also be viewed as a special caseof the general weighted index with weights reflecting relativeimportance of subset of subjects. However, while selection ofweights for the general index could be arbitrary, weights ofsubject-centered FROC curve are determined by the a prioriknown cluster size, and consequently, has an objective prob-abilistic interpretation. Using the derived formulation of thesubject-centered FROC curve, many conventional summaryindices could be derived in a straightforward manner.

We discussed several important properties of the subject-centered approach. In scenarios where the association be-tween the number of targets and performance characteristicsis not substantial, the estimates of subject-centered quantitiescould be numerically similar to their target-centered coun-terparts. In these scenarios, the subject-centered and target-centered differences between modalities are likely to be inthe same direction and of similar magnitude; and a lower sta-tistical power of subject-centered techniques can make themless attractive than conventional target-centered approaches.When statistical power becomes an issue, the subject-centeredapproach could be used for verifying robustness of theinferences.

The approach presented in this paper combines FROCcurves corresponding to groups of subjects with differentnumber of targets by averaging the operating points corre-sponding to the same rating threshold. In practice, the deci-sion threshold regarding a subject could naturally depend onthe number of known targets (which often impacts the numberof reported findings). Hence, it would be natural to combinethe group-specific curves by averaging the points correspond-ing to the same decision threshold, which would most likelycorrespond to different ratings in different groups. Thus, anatural extension of the approach presented here would beto consider vertical averaging as well as other practically rea-sonable types of calibration approaches. Another practicallyuseful direction for future research would be to develop anapproach for the analysis of multireader studies, which arecommonly employed in the evaluation of detection and local-ization systems.

The developed approach most directly impacts diagnostictest evaluation field. Assessment of diagnostic performanceis increasingly more focused on the detection and localizationtechniques used for identifications of multiple targets per sub-ject (e.g., nodules in the lungs, affected lymph nodes, massesin the breast). The analytical method we develop helps ad-dress a frequently relevant question that cannot be answeredby the conventional FROC curve and helps avoid erroneousimpression of the superiority of practically suboptimal diag-nostic techniques. As such, this method can improve multi-ple stages of development, optimization, evaluation, regula-tory approval, and practical utilization of diagnostic systemsby helping identify diagnostic systems that are beneficial intasks where patient-centered characteristics are important.

ACKNOWLEDGMENT

This research was in part supported by the National Insti-tute of General Medical Science of the National Institutes ofHealth under Award No. R01GM098253.

a)Author to whom correspondence should be addressed. Electronic mail:anb61@pitt.edu; Telephone: +00-412-383-5738; Fax: +00-412-624-2183.

1J. P. Egan, Signal Detection Theory and ROC Analysis (Academic, NewYork, 1975).

2J. A. Swets and R. M. Picket, Evaluation of Diagnostic Systems: Methodsfrom Signal Detection Theory (Academic, New York, 1982).

Medical Physics, Vol. 40, No. 5, May 2013

Bandos et al. (2013), Medical Physics 40:051706-9

Two major themes were highlighted during the interviews aboutthe delivery experiences including thoughts and perception ofdeath and dealing with gaps in memory.

Thoughts and perception of death. Many women men-tioned death repeatedly throughout their interviews and discussedthe impression that death was close and imminent and in somecases thinking that they were already dead.

‘‘I thought I would die… I could die at that time. Especially when thedoctor showed the dress he was wearing… Where I was lying, I sawthat that doctor’s dress was soaked with blood. And then he squeezed, isit gauze or something, and I saw the blood. I was afraid at that timethat I would die.’’ (37 years old, near miss, live birth)

‘‘I thought I was dead, but when I became conscious I realized that Iwasn’t feeling the abdominal pains any longer, and there was a plasteron my belly.’’ (35 years old, near miss, stillbirth)

Table 1. Criteria to identify potentially life-threatening conditions and near miss [16].

POTENTIALLY LIFE-THREATENING CONDITIONS

Severe complications

1. Severe postpartum hemorrhage: genital bleeding after delivery, with at least one of the following perceived abnormal bleeding (1000 mL or more) or any bleedingwith hypotension or blood transfusion.

2. Severe preeclampsia: Persistent systolic blood pressure of 160 mmHg or more or a diastolic blood pressure of 110 mmHg; proteinuria of 5 g or more in 24 hours;oliguria of ,400 ml in 24 hours; and HELLP syndrome or pulmonary edema. Excludes eclampsia.

3. Eclampsia: generalized fits in a patient without previous history of epilepsy. Includes coma in preeclampsia.

4. Sepsis or severe systemic infection: presence of fever (body temperature .38uC), a confirmed or suspected infection (e.g. chorioamnionitis, septic abortion,endometritis, pneumonia), and at least one of the following- heart rate.100, respiratory rate.20, leukopenia (white blood cells ,4000), leukocytosis (white blood cells.12 000)

5. Ruptured uterus: ruptured uterus during labour

Critical Interventions

1. Use of blood products

2. Laparotomy (including hysterectomy, excluding C-section)

3. Admission to Intensive Care Unit/recovery room .= 6 hours

NEAR-MISS CRITERIA

Clinical Organ Dysfunction

1. Acute cyanosis

2. Gasping

3. Respiratory rate .40 or ,6 bpm

4. Shock

5. Cardiac Arrest

6. Oliguria non-responsive to fluids or diuretics

7. Any loss of consciousness lasting .12 hours

8. Stroke

9. Uncontrollable fit/status epilepticus

10. Global paralysis

11. Jaundice in the presence of pre-eclampsia

Laboratory markers of organ dysfunction

12. O2 saturation ,90% for more than 60 min

13. PaO2/FiO2,200 mmHg

14. Creatinine.300umol/ml or .3.5 mg/dL

15. Bilirubin.100umol/L or .6.0 mg/dL

16. pH,7.1

17. Lactate .5mEq/L

18. Acute thrombocytopenia (,50,000 platelets)

Management-based proxies

19. Hysterectomy following infection of hemorrhage

20. Use of continuous vasoactive drugs

21. Cardio-pulmonary resuscitation

22. Dialysis for acute renal failure

23. Any non-anesthetic intubation or ventilation

24. Transfusion of .5 units of blood or red cells

doi:10.1371/journal.pone.0044536.t001

Severe Maternal Morbidity and Quality of Care

PLOS ONE | www.plosone.org 4 August 2012 | Volume 7 | Issue 8 | e44536

Tunçalp et al. (2012), PLoS ONE 7(8):e44536

Sensors 2013, 13 5331

Figure 7. Effect of choice of wavelet.

As with Experiment 1, classifiers are very competent at identifying inertial movement activities(A1–A4), but game activities (A5–A7) pose more of a challenge. The inertial activities are very distinctas the energy during these activities is unique. They range from zero energy output when the player isstationary to maximum energy when the player is sprinting. The confusion encountered between thegame activities is due to the similar motions being performed. In soccer these motions involve lowerleg movement while in hockey these game activities involve the upper arm movement. Table 5 providesthe parameters for the highest classification accuracy attained for each respective sport. Tables 6 and 7display their confusion matrix data. It took 5 ms for this approach to classify the extracted DWT features.

Table 5. Highest classification accuracies attained for Experiment 2.

Device Sport Classifier DWT lvl Mother W. Length (Sec) F-Measure

Smartphone Soccer NaiveBayes 6 rbio1.1 3 0.799Smartphone Hockey MLP 6 bior1.1 7 0.823

Table 6. Confusion matrix for Football Smartphone data for Experiment 2.

Activity A1 A2 A3 A4 A5 A6 A7

A1 28 0 0 0 0 0 0A2 0 30 0 0 0 0 0A3 0 0 30 0 0 0 0A4 0 0 0 30 0 0 0A5 0 1 0 0 24 4 1A6 0 2 0 0 9 12 7A7 0 1 0 0 12 2 15

Mitchell et al. (2013), Sensors 13:5317-5337

Figuras retiradas de: http://abacus.bates.edu/~ganderso/biology/resources/writing/HTWtablefigs.html

Copelli et al. (2002), Physical Review E 65:060901(R) denoting some kind of memory effect, a phenomenon ob-served previously by Chialvo et al. !8" and Lewis and Rinzel!9".Due to a chain-reaction mechanism, the spike of a single

receptor cell is able to excite all the other cells. The sensi-tivity per neuron has thus increased by a factor of L. This canbe clearly seen in Fig. 2, which shows the average firing rateper cell F in the coupled system #top panels$, as well as theamplification factor A%F/ f #bottom panels$. This is a some-what expected effect of the coupling: neuron j is excited bysignal events that arrive not only at neuron j but elsewhere inthe network.More surprising is the fact that the dynamic range #the

interval of rates where the neuron produces an appreciablebut still nonsaturating response$ also increases dramatically.This occurs due to a second effect, which we call the self-limited amplification effect. Remember that a single spike ofsome neuron produces a total of L neuronal responses. Thisis valid for small rates, where inputs are isolated in timefrom each other. However, for higher signal rates, a newevent occurs at neuron k before the wave produced by neuronj has disappeared. If the initiation site k is inside the fronts ofthe previous wave !e.g., the events signaled by arrows in Fig.1#b$", then two events produce 2L responses as before. But ifk is situated outside the fronts of the j-initiated wave !as inthe first input events shown in Fig. 1#b$", one of its frontswill run toward the j-wave and both fronts will annihilate.Thus, two events in the array have produced only L exci-

tations #that is, an average of L/2 per input event$. So, in thiscase, the efficiency for two consecutive events #within a win-dow defined by the wave velocity and the size L of the array$has been decreased by half. If more events #say, m) arriveduring a time window, many fronts coexist but the averageamplification of these m events #how many neurons eachevent excites$ is only of order L/m .Therefore, although the amplification for small rates is

very high, saturation is avoided due to the fact that the am-plification factor decreases with the rate in a self-organized

nonlinear way. The amplification factor A shown in Figs.2#c$ and 2#d$ decreases in a sigmoidal way from A!O(L)for very small rates #since a single event produces a globalwave$ to A!1 for large rates, where each cell responds as ifisolated since waves have no time to be created or propagate.The role of the system size L for low input rates becomes

clear in Fig. 2#c$: the larger the system, the lower the rate rhas to be in order for the amplification factor to saturate atO(L). In other words, we can think of a decreasing crossovervalue r1(L) such that the response is well approximated byF(r)!L f (r)&Lr for r"r1(L). In this linear regime con-secutive events essentially do not interact. Larger systemsizes increase not only the overall rate of wave creation!'1#(1#r)L" but also the time it takes for a wave to reachthe borders and disappear. In the opposite limit of large inputrates, the behavior of the response is controlled by the abso-lute refractory period ( , as shown in Fig. 2: F and f saturateat r2%1/( for f$r2, independently of the system size.So what happens for intermediate input rates, i.e., r1"r

%r2? The answer is a slow, Weber-Fechner-like increase inthe response F, as can be seen in Fig. 3. The logarithmicdependence on r is a good fit of the curves for about threedecades.Motivated by results obtained with more realistic ele-

ments !6" we introduced a relative refractory period in ourCA model. We first define a time window M after a spikeduring which no further spikes can occur #absolute refractoryperiod$. In the following n#M#2 steps #relative refractoryperiod$, a single input does not produce a spike but two ormore inputs can elicit a cell spike if they arrive within atemporal summation window ) #details of this model will bedescribed in a forthcoming full paper$. This ingredient pro-duced the appearance of a power law F(r) curve #Stevenslaw !1,2"$, as shown in Fig. 4. Notice that the exponent de-pends on the relative refractory period. The appearence of apower law transfer function is a robust effect also observedin coupled maps systems !6".

FIG. 3. F&( vs input rate r for L!5000 #open symbols$ andL!200 #filled symbols$ for different values of n. A L!25000 curvefor n!50 #crosses$ shows no difference to the L!5000 case.Straight lines are intended as a guide to the eye. Inset: F(r) for theHodgkin-Huxley system.

FIG. 4. Neuronal ‘‘Stevens law’’ F*r+ in automata which takestemporal summation effects into account #see text for details$. Fir-ing rate F vs input rate r for a CA with n states and an absoluterefractory period of M!3 time steps. Filled circles: n!15, )!10, +!0.38; open circles: n!100, )!80, +!0.44.

RAPID COMMUNICATIONS

PHYSICS OF PSYCHOPHYSICS: STEVENS AND . . . PHYSICAL REVIEW E 65 060901#R$

060901-3

Figura inserida dentro da figura (inset)

Shenvi et al. (2011), J. Chem. Phys. 134:144102

Inset

lines of evidence that fusion occurs preferentially in obligatecomplexes, including a lower tendency for fusing subunits tobe observed in isolation and a much higher propensity for corre-latedmessenger RNA (mRNA) expression. Importantly, we showthat the observed assembly conservation does not arise from atendency for fusion to occur in obligate complexes.Taken together, our results provide robust evidence of evolu-

tionary selection for assembly-conserving gene fusion events.Importantly, we emphasize that this is not an absolute rule,and that a slight majority of fusions do in fact disrupt assembly.However, one must consider that random subunit fusions wouldconserve (dis)assembly in only a very small fraction of cases andthus the evolutionary frequency of (dis)assembly-conserving fu-sions is far higher than would be expected by chance.

Optimization of Assembly upon Fusion throughSimplification of Protein Complex TopologiesDespite the strong selection for assembly conservation, it is clearthat many evolutionary fusion events have modulated existingassembly pathways. Thus, we hypothesized that there mayhave been further evolutionary selection for fusion events thatoptimize assembly. For instance, although any fusion event be-tween subunits will reduce the number of assembly steps by atleast one, greater simplification will occur if the fusion involvestwo subunits that both share other interaction partners, as thiswill result in fewer intermolecular interfaces in the fused complex(Figure 4A).We compared the reduction of intersubunit interfaces in pro-

tein complexes upon fusion with what would be expected iffusion occurred randomly between subunits (essentially as inFigure 3C). Interestingly, we observed that gene fusion eventstended to reduce the number of interfaces by considerablymore than would be expected by chance (2.90 versus 2.21,p = 13 10!4; Figure 4B). This strongly implies evolutionary selec-tion for fusions that maximally reduce the number of interfaces in

a protein complex, thereby simplifying their topologies and as-sembly pathways. We suggest that having fewer intersubunit in-terfaces would both lower the risk of misassembly and increasethe speed of assembly.We investigated this phenomenon further by searching high-

throughput interaction data for interacting proteins with evi-dence of fusion occurring between them. Each binding partnershared by a pair of proteins will further reduce the number ofdistinct protein-protein interactions by one upon fusion (Fig-ure 4C). Pairs of proteins from Escherichia coli that undergofusion share a mean of 19.2% of their binding partners,compared with 13.2% expected for random fusions within theinteraction network (p = 3 3 10!4; Figure 4D). Similar trendsare also seen in yeast (14.7% versus 7.1%, p = 0.008), humans(23.2% versus 16.4%, p = 0.04), and a large number of otherspecies (Table S3). Contrary to our structure-based analysis, iftwo proteins share a binding partner in these high-throughputdata, it does not necessarily mean that they are interacting simul-taneously (Kim et al., 2006a). Nevertheless, these results implyevolutionary selection for fusion events that optimize network to-pology by reducing the number of discrete protein interactions,in analogy to the simplification of assembly.

Protein Structural Constraints on FusionBecause gene fusion essentially forces a pair of proteins tointeract permanently with each other, the influence of fusion onassembly may be limited by protein structural constraintsdictating whether or not a fusion event is likely to occur. Uponfusion of two proteins, the C terminus of the first will becomecovalently linked to the N terminus of the second. If these terminiare far apart in the prefusion complex, fusionwould require eitherthe addition of a lengthy linker or a major disruption of the inter-subunit interface. However, if these termini are close in space,fusion would be more likely to conserve the existing quaternarystructure (Figure 5A).

Figure 3. Evolutionary Conservation of Protein Complex (Dis)Assembly Pathways upon Gene Fusion(A) Comparison of the frequency of evolutionary gene fusion events in heteromeric subunits pairs that would either conserve or modify (dis)assembly pathways

upon hypothetical subunit fusion.

(B) Comparison of observed (dis)assembly conservation from in vitro experiments and in silico predictions with the intrinsically expected values for complexes

with the same topologies.

(C) Direct comparison of predicted (dis)assembly conservation and randomly occurring fusions in complexes with more than two unique subunits. Error bars

represent the SEM.

See also Figure S3 and Table S1.

Cell 153, 461–470, April 11, 2013 ª2013 Elsevier Inc. 465

Marsh et al. (2013), Cell 153:461-470

Figura composta. Quando existem gráficos e/ou outros tipos de ilustrações inter-relacionados, pode ser mais eficiente construir uma figura composta a partir deles. Uma figura composta combina vários gráficos e/ou outras ilustrações e uma só figura com uma única legenda. Cada figura e ilustração deve ser indicada claramente por uma letra (maiúscula ou minúscula dependendo da revista ) e, quando referida no texto, deve ser identificada por essa letra, p. ex. (Fig. 3B).

oftenmore flexible in isolation and tend to undergo larger confor-mational changes upon binding (Marsh and Teichmann, 2011;Marsh et al., 2012). Furthermore, the presence of multipledistinct subunits means that heteromers have far more potentialroutes of assembly, which could complicate predictions.To test the association between interface size and assembly,

we performed nanoelectrospray ionization (nESI)-MS experi-ments (Sobott et al., 2002; Hernandez and Robinson, 2007) onfive of the prefusion complexes identified above in order todetermine their reversible in vitro disassembly pathways. Repre-sentative mass spectra are shown in Figures 2A and S1.

Although the process of disassembly is different from that of as-sembly, the two processes are generally reversible in homomericcomplexes (Levy et al., 2008). To further support this notion, weshow that the prefusion complexes studied here can be reas-sembled from their dissociated states without the formation ofoff-pathway subcomplexes, thus demonstrating the reversibilityof assembly and disassembly in heteromers (Figure S2). There-fore, we refer to ‘‘(dis)assembly’’ as this reversible process wecan probe in solution.In addition to the MS experiments, we also identified four pre-

fusion complexes in which (dis)assembly pathways could be

Figure 2. Experimentally Characterized (Dis)Assembly Pathways of Heteromeric Prefusion Complexes(A) (Dis)assembly pathways of complexes characterized by nESI-MS aswell as representativemass spectra. See Table S2 for a full list of subcomplexes identified

under different solution conditions.

(B) (Dis)assembly pathways of complexes identified from previously published experiments. In the graph representations of protein complexes, interfaces that

undergo fusion are shown in orange.

See also Figure S1.

Cell 153, 461–470, April 11, 2013 ª2013 Elsevier Inc. 463

To illustrate this, we consider the case of the prefusion com-plex Klebsiella aerogenes urease (Jabri and Karplus, 1996),where fusion is known to occur between genes correspondingto the g and b subunits. Because the g subunit fuses upstreamof the b subunit, fusion will result in a linkage between the C ter-minus of the g subunit and the N terminus of the b subunit. Exam-ination of the complex crystal structure reveals that these terminiare in fact quite close, separated by only 16 A (Figure 5B). Wewillrefer to this as the ‘‘fusion distance.’’ The ‘‘reverse distance’’ (iffusion were to occur in the opposite gene order [i.e., b upstreamof g]) is much greater (66 A).

We systematically compared the fusion and reverse dis-tances of all prefusion complexes in our data set in which thesubunits correspond closely to the full-length genes (Figure 5C).We observe that for cases in which fusion has occurred in only asingle gene order, the fusion distances are shorter than thereverse distances in 35/47 (74.5%) fusion events (p = 0.001,binomial test). Furthermore, the mean fusion distance is14.1 A shorter than the mean reverse distance (p = 0.001, Wil-coxon signed-rank test). Importantly, this tendency for fusionto occur between the closer termini is not related to the (dis)as-sembly conservation demonstrated earlier (see ExtendedExperimental Procedures). Therefore, the order of gene fusionis closely related to the structure of protein complexes, with sig-nificant evolutionary selection for fusion events that link moreproximal termini. This is consistent with a previous study inwhich pairs of domains that were observed to interact both in-ter- and intramolecularly, which included several fusions, wereshown to conserve their binding orientations in most cases(Kim et al., 2006b).

Figure 4. Evolutionary Simplification of Pro-tein Complex Assembly via Gene Fusion(A) Graph representation of a prefusion complex

(PDB ID: 1RM6) in which the subunits that fuse (a

and g) share interaction partners, leading to a large

decrease in the number of interfaces upon fusion.

(B) Mean reduction in interfaces (per protomer)

upon fusion for 36 fusion events, compared with

random fusions within the same complexes.

(C) Protein-protein interaction network for the

E. coli proteins cysI and cysJ showing that four out

of nine binding partners (magenta) are shared

between the two; thus, the total number of discrete

interactions will be reduced by four upon fusion.

(D) Comparison of shared binding partners be-

tween proteins that undergo fusion from high-

throughput protein interaction data for E. coli (n =

61), yeast (n = 16), and humans (n = 16). Com-

parisons for 411 other species are provided in

Table S3. Error bars represent the SEM.

See also Table S1.

DISCUSSION

By comparing the identities of assemblyintermediates observed in nESI-MS ex-periments with the structures of proteincomplexes, we were able to gain a funda-

mental mechanistic insight into protein assembly. Essentially,assembly in both homomeric and heteromeric complexes isdriven by the hierarchy of interface sizes within a protein com-plex, such that assembly intermediates will tend to possesslarger intersubunit interfaces. By taking advantage of Nature’sgrand protein engineering experiment, i.e., the large number ofgene fusion events that have occurred throughout evolutionaryhistory, we show that these assembly intermediates are underevolutionary selection. This suggests that modifying existing as-sembly pathways has a significant tendency to lower an organ-ism’s evolutionary fitness.Although numerous functional benefits arise from the forma-

tion of multisubunit complexes, the increased complexity isassociated with a greater risk of misassembly. Our results sug-gest that evolution has selected for protein complexes thatassemble via well-defined, ordered pathways. Presumably, thisleads to faster and more efficient formation of the functionalcomplexes. If these assembly pathways become modified inevolution, the identities of the assembly intermediates willchange, potentially increasing their susceptibility to misassem-bly or aggregation. Thus, the evolutionary conservation and opti-mization of assembly pathways revealed here provide a potentialmeans of minimizing these risks while maintaining the advan-tages of complex formation. Furthermore, our results have prac-tical implications in that the identities of assembly intermediatescan now be predicted from the three-dimensional structures ofprotein complexes. This may provide clues as to howmisassem-bly occurs and how it might be prevented.The assembly and quaternary structure of protein complexes

are highly important for determining which gene fusion events

466 Cell 153, 461–470, April 11, 2013 ª2013 Elsevier Inc.

are selected. Since the vast majority of hypothetical fusionevents would modify existing assembly pathways, this helps torationalize why most protein interactions are not predicted byfusion-basedmethods (e.g., only 3.7%of the nonredundant sub-unit pairs in our data set are associated with evolutionary fusionevents). In addition, we demonstrated further selective pressureupon fusion related to assembly optimization and the require-ment for covalent linkage of termini.These findings provide amore detailed, structural understand-

ing of fusion that should allow one to better interpret and utilizefusion-based predictions. Furthermore, fusion-based strategieshave been gaining prominence in the field of protein engineering(Padilla et al., 2001; Sinclair et al., 2011; Lai et al., 2012). Our in-sights can also potentially guide future protein engineeringapproaches: if covalent fusion of subunits is desired in order tostabilize a complex, success is most likely to be achieved withengineered fusions that conserve existing assembly pathwaysand in which the gene order is chosen to best match the existingquaternary structure.This work also reveals an evolutionary connection between

protein and genome structure. In 13% of the cases we exam-ined, fusion occurred in both orders (i.e., AB and BA), in similarityto previous work showing that the vast majority (!92%) ofdomain pairs occur in only a single order (Apic et al., 2001). Ithas been suggested that the order of domain combinations inmultidomain proteins is due primarily to historical chance, asdomain pairs with the same structure and function can occur inboth orders given the presence of a long interdomain linker(Bashton and Chothia, 2002; Vogel et al., 2004). Thus, multido-main proteins are highly versatile and a short interterminal fusiondistance is not a strict requirement. However, our results suggestthat the formation of a long linker (as required to preserve thequaternary interaction) can be a limiting factor, because weobserve a strong preference for fusions in the order correspond-ing to the shorter interterminal distance. Therefore, our work im-plies that, rather than being an evolutionary artifact, the order in

which genes fuse can be directly related to the structural fea-tures of the proteins they encode, thus demonstrating a simpleway in which protein structure can influence genomicorganization.Finally, our results highlight a fascinating connection between

evolutionary processes, which act over millions of years, and as-sembly, which occurs on the order of seconds. Although the as-sembly pathways of homomeric complexes were previouslyfound to reflect their evolutionary histories (Levy et al., 2008),here we observed an opposite phenomenon in which the evolu-tionary process of gene fusionmimics heteromer assembly in or-der to conserve the existing assembly pathway.

EXPERIMENTAL PROCEDURES

Structural Data SetsWe started with the full set of heteromeric biological units from protein crystal

structures in the RCSB Protein Data Bank (Berman et al., 2000). We filtered

heteromers formed by polypeptide cleavage by identifying different chains

with the same external database reference identifier (db_id, which generally

corresponds to the UniProt sequence) but with a sequence identity of

<90%. Only subunits with at least 50 residues were considered. Protein com-

plexes containing nucleic acids were ignored because we have no way of reli-

ably predicting (dis)assembly for these cases.

We filtered subunit pairs from the protein complexes for redundancy, first by

grouping them by their SUPERFAMILY domain assignments (Gough et al.,

2001) and then by calculating the sequence identities between all pairs in

each group. If both subunits from a pair had >70% sequence identity to

another pair, only the pair from the higher-resolution crystal structure was

kept. After the sequence redundancy filtering was completed, we had a total

of 2,544 nonredundant heteromeric subunit pairs. All subunit pairs used in

this study, along with their various relevant properties, are provided in

Table S1.

For each complex, we calculated the size of the interfaces between all pairs

of subunits using AREAIMOL (Collaborative Computational Project, Number 4,

1994). In complexes containing more than one copy of each subunit, there can

be more than one interface for a given pair of subunit types (e.g., the two

different a-b interfaces in 2F9Y; see Figure 2A). Therefore, in compiling our

nonredundant set of subunit pairs, we only considered the largest interface

Figure 5. Protein Structural Determinants of Gene Fusion(A) Fusion may be unable to occur if the protein termini are too far apart in the prefusion complex. However, if the C terminus of one subunit is close to the N

terminus of the other, a productive fusion is more likely.

(B) Comparison of fusion and reverse distances between the g and b subunits of K. aerogenes urease (PDB ID: 1KRA; only one abg trimer from the full (abg)3nonamer is shown).

(C) Box plot comparison of fusion and reverse distances (in A) in 47 fusion events from full-length proteins in which fusion occurs in only a single gene order; black

bars represent the medians, and boxes and whiskers indicate the distribution quartiles.

See also Table S4.

Cell 153, 461–470, April 11, 2013 ª2013 Elsevier Inc. 467

Exemplos de figuras compostas Marsh et al. (2013), Cell 153:461-470

Evans et al. (2011), Med. Phys. 38:1448-1458

Evans et al. (2011), Med. Phys. 38:1448-1458

Garcia-Perez et al. (2005), J. Neurosci. 25:2597-2608

Outros exemplos de ilustrações

model remains 60 mm by 60 mm by 30 mm and the length of eachdipole remains 29 mm with a 0.5-mm gap in the middle. Thesubstrates for the front, back, and circuit boards are 1 mm-thickFR4 with relative permittivity 4.6 and loss tangent 0.02. The di-rections of the amplifier MGA53543 are marked by the black ar-rows. The lumped elements mounted on the circuit board aremarked in different colors. C1 ¼ 100 pF, C2 ¼ 10 nF, andL1 ¼ L2 ¼ 27 nH. Simulation performed at 885.45 MHz for avertically polarized incident wave is shown in Fig. 3C. Alongthe direction from the source to the receiver, the polarization an-gle rotates by þ16.75 degrees and −16.75 degrees for front-sideincidence and back-side incidence, respectively. The dc voltage ofthe amplifiers for these results is 0.533 V.

Finally, the Faraday-like rotation of the proposed gyrotropicmetamaterial is confirmed in a transmission measurement experi-

ment. The photo of a metamaterial sample is shown in Fig. 4A,with the top-layer circuit board seen in Fig. 4B. Blue and redarrows indicate the directions of the amplifiers. A two-ridge hornantenna, serving as the source, is polarized along the vertical di-rection and driven with power at 10 dBm by a Hewlett-Packard8350B Sweep Oscillator. A 133 mm-length printed dipole anten-na on a 1 mm-thick FR4 substrate serves as a detector on theother side of the sample and is rotated to measure the polariza-tion of the field through a Hewlett-Packard 8756A Scalar Net-work Analyzer. We first measured the polarization change withthe wave incident from the front side to the back side of the sam-ple, then flipped the sample horizontally and measured the po-larization change with the wave incident from the back side to thefront side. Along the direction from the source to the detector, weidentify the change in the polarization angles, at a dc voltage of

L1 & L2C1 C2Amplifier

DC &GND

B

30

210

60

240

90

270

120

300

150

330

180 0 0.2 0.4 0.6 0.8 1

C

PowerF to BB to F

PolarizationF to BB to F

L2

L2

L2

L2

C1 C2 L1DC

IN

OUT

A

GND

GND

Fig. 3. Gyrotropic metamaterial for vertical polarization and its simulation results. (A) The circuit schematic of a basis assembly. (B) The structure schematic of aunit cell. (C) The simulated polarizations (solid) and power patterns (dashed) at the receiver when a vertically polarized wave at 885.45 MHz is incident in twoopposite directions.

940 942 944 946 948 95045

60

75

90

105

120

135

Frequency (MHz)

Ang

le o

f Max

. Rec

eive

d P

ower

F to BB to F

CA

D

Back

Front

B30

210

60

240

90

270

120

300

150

330

180 0F to BB to F 0.2 0.4 0.6 0.8 1

0.1λ 0.1λ

0.1λ

Front

Fig. 4. Photos of the single-polarization gyrotropic metamaterial sample and its experimental results. (A) The photo of a single-layer gyrotropic metamaterialfor vertical polarization. (B) The photo of the circuit board in the sample. (B, Inset) Photos of themetallic pattern on the front and the back board of the sample.(C) Angles at which the received power is maximized for a given incident frequency. (D) Measured power patterns calibrated from the free-space measurementwith a vertically polarized wave incident at 946 MHz.

13196 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1210923109 Wang et al.

Wang et al. (2012), PNAS 109:13194-13197

Resumográfico(graphicalabstract)•  Algumasrevistaspedemqueoartigosejaacompanhadode

umresumográfico.•  Normalmente,oresumográficoémostradonasversões

onlinedoartigoedoíndicedovolumedarevista(volumecontents).

•  Oresumográficoéumaúnicafigura(simplesoucomposta)quedescrevedeformavisual,pictóricaeconcisaaprincipalmensagemdoartigo.

•  Afiguracorrespondenteaoresumográficodevesersubmetidacomoumafiguraseparadanoprocessodesubmissão.

•  VejaapáginadaEditoraElsevierparaexemplosderesumosgráficos:http://www.elsevier.com/authors/graphical-abstract

Exemplo da versão online de um artigo recente na revista Cell. Note que além do resumo gráfico e do resumo (summary), há um campo para highlights.

Exercício

•  Selecioneasrevistascientíficasmaisimportantesdasuaárea,ouaquelasemquevocêgostariadepublicarumartigo;

•  Leiaalgunsartigosdessasrevistas;•  Presteatençãonastabelasefigurasdosartigos;

•  Tenteverificarseelasestãomaisoumenosdeacordocomassugestõesdadasnestaaula.

Aescolhadarevista

•  Aescolhadarevistarequerumaanálisecuidadosa.

•  Háváriosfatoresaseconsiderarnahoradeescolherarevista.

•  Arevistatemrevisãoporpares?•  Arevistapublicaatualmenteartigoscomooseu?

•  Seestivesseprocurandoumartigocomooseu,emquerevistaoprocuraria?

•  Quaissãoasrevistasdemelhorreputaçãonasuaárea?

•  Pergunteacolegasquerevistaselesrespeitam.

•  Dêumaolhadanosartigosrecentesdarevistaejulguesuaimportância.

•  Ocorpoeditorialécompostoporlíderesnosseuscampos?

•  Qualofatordeimpactodarevista?•  Quaissãoasrevistasmaisprováveisdeseremcitadaspelospesquisadoresnasuaárea?

•  Arevistaépublicadaporumasociedade?•  Asrevistasdesociedadessão,emgeral,asdemaiorprestígioedemaiorcirculação.

•  Tenhacuidadocomasrevistasnovas(impressasoudeacessolivrepelainternet),especialmentecomaquelasnãopatrocinadasporumasociedade.

•  Arevistaéindexadapelasprincipaisbasesdedadoseletrônicas:Medline,BiologicalAbstracts,ChemicalAbstracts,CurrentContents,etc?

•  Existemrevistascujosleitoresvocêprecisa/querinfluenciar?

•  Querevistaspoderiamdarum“tratamentojusto”aoseuartigo?

•  Qualaperiodicidadedarevista?•  Qualotempomédioentreadataderecebimentodeumartigoeadatadesuapublicação?

•  Usandoasinformaçõesdadasem“datesubmitted”e“dateaccepted”ecomadatadepublicaçãodoartigoemmãossepodeestimaraduraçãodoprocessoderevisãoeotempotranscorridoentreaaceitaçãoeapublicaçãodefato.

•  Emquelíngua(s)sãopublicadososartigosdarevista?

•  Qualofocodarevista?– Éamploourestrito?– Quedisciplinasdoconhecimentoestãorepresentadas?

– Qualaorientaçãodepesquisadarevista:•  Básica;•  Teórica;•  Aplicada?

•  Vocêgostadaaparênciadosartigospublicados(formato,tipodeletraeestilousadoparacitarasreferências)?

•  Casorelevante,arevistapublica“shorte/ourapidcommunications”?

•  Asfiguraspublicadaspelarevistatêmaresoluçãoquevocêprecisa?

•  Arapidezdepublicaçãoéimportante?•  Casopositivo,revistasmensais(monthly)saemcommaisfrequênciaquerevistastrimestrais(quarterly).

•  Casoaescolharecaiasobreumarevistadeacessolivre(openaccess),quecobradosautorespelapublicação:vocêouoseugrupo/instituição/agênciadefomentoteriaodinheironecessárioparapagar?

Respondendoparaosrevisores

•  Apósasubmissãodoartigo,oautorcorrespondentereceberáumacartadoeditordarevistacomoscomentáriosdosrevisores,cujasidentidadessãomantidasemsegredo.

•  Acartaindicaráumadeduascoisas:– Rejeiçãodoartigo;ou– Aceitaçãoprovisóriadoartigo.

Artigorejeitado

•  Seoeditorrejeitaroartigo,acartadeverádarasrazõesparaessadecisão.

•  Nestecaso,osautoresdevemleroscomentáriosdosrevisoresparadeterminarseoartigopoderiaseraceitoseforemfeitascertasmudanças.

•  Nãosedevetomaroscomentáriosdosrevisorescomocríticaspessoais.

•  Namaioriadoscasos,oeditoreosrevisoresestarãotentandoajudarosautoresaproduzirumartigodemelhorqualidade.

•  Emalgunscasos,arejeiçãopodeserdevidaaummomentoinadequadodesubmissão:arevistapodeteracabadodeaceitaroudepublicarumtrabalhosimilar.

•  Sempresepodesubmeteroartigoaumaoutrarevista.

•  Nestecaso,ésemprebomlevaremconsideraçãooscomentáriosdosrevisores.

•  Mesmoqueosautoressintamqueosrevisoresnãoentenderamalgonoartigo,outrosrevisorestambémpoderãonãoentender.

•  Seoeditorachaqueoassuntodoartigonãoseenquadranoescopodarevista,nãoháporquelutarcontraisso.

•  Nestecaso,nãoháoutraescolhaanãosersubmeteroartigoaoutrarevista.

•  Asubmissãoaoutrarevistadeveserfeitarapidamente.Algunsdadospodemperderrelevânciaselevaremmuitotempoparaserpublicados.

Artigoaceitoprovisoriamente

•  Nestecaso,osautoresprecisamplanejarumaestratégiaderevisãodoartigoparaqueelesejaaceitoemdefinitivo.

•  Istoincluiráaresubmissãodoartigorevisadoedeumacartacomasrespostasaoscomentáriosdosrevisores.

•  Asseguintessugestõespodemajudarnarevisão:

•  Deve-selertodososcomentáriosdosrevisoresedoeditor.

•  Nuncasedeveresponderimediatamente.Deve-sepermitiralgunsdiasparareflexãosobreoscomentários.

•  Seoscomentáriosdosrevisoresedoeditorpuderemserusadosparamelhoraroartigo,asmudançassugeridasporelesdevemserfeitas.

•  Depoisdosdiasdereflexão,nãosedeveperdertempoemresponder.

•  Osautoresdevemcomeçararascunharumarespostaclara,detalhada,pensadaeeducada.

•  Deve-seevitarumtomdefensivooudeconfrontaçãonarespostaaosrevisores.

•  Osautoresdevemextrairinformaçõesúteisdoscomentários,aceitarassugestõesquemelhoremoartigoe,calmamente,explicaroseupontodevistaquandoelediferirdodosrevisores.

•  Deve-serespondercompletamenteacadacomentárionaordememqueelesforamfeitos.

•  Asrespostasdevemserenumeradas.•  Senecessário,deve-secopiarecolarnacartaostrechosdoartigoquemudarammuito.

•  Nãohálimitedepáginasparaacartaderespostadosautores.

•  Amaioriadoseditoresgostaderespostaslongasecompletas.

•  Deve-sealteraroartigoondeseacharqueasmudançassugeridasfazemsentido.

•  Osautoresnãosãoobrigadosafazertodasasmudançassugeridas,maselestêmqueresponderatodososcomentários.

•  Seumasugestãoforrejeitada,oeditoriráquererumaboarazãoparaissocomevidênciasapoiadaspelasreferências.

•  Dizerapenasqueosautorespreferemdojeitoqueestánãoéumaboarazão.

•  Osrevisoresnemsempreconcordamunscomosoutros.

•  Nestecaso,osautoresdevemfazerumaescolha.

•  Devemdecidirquaisdasrecomendaçõesparecemmaisválidasedevemescreveremsuarespostaaoeditorqueosrevisoresfizeramsugestõesconflitantesequeosautoresescolheramaquelaquelhespareceamelhorsugestão.

•  Seficaróbvioqueumrevisorcometeuumerro,deve-seargumentarcontraasugestãodeleefornecerevidênciasparamostraroerro.

•  Algumasvezesosrevisoresouoeditorpedemqueotamanhodoartigosejareduzidoconsideravelmente.

•  Nestecaso,osautoresnãodevemsesentirtãoapegadosàspalavrasqueescreveramedevemencurtaroartigo.

•  Osautoresdevemsecertificarqueoqueelesdisseramnacartaquefoimudadonoartigotenhasidodefatomudado,etambémqueoartigomodificadocontinuadeacordocomasguidelinesdarevista.

•  Oseditoresficamirritadosquandodescobremqueoscomentáriosfeitosnacartaderespostanãoconcordamcomoqueestánoartigo.

Apublicaçãodeumartigodevesercelebrada

•  Oprocessoquelevaàpublicaçãodeumartigoemumarevistacientíficacomrevisãoporpareséumdesafio,masnofim,depoisquetodootrabalhofoifeitoeoartigofoipublicado,érecompensadorveroartigonarevista.

•  Issomerecesercelebrado!

Razõespelasquaisumartigoérejeitado

•  Háváriasrazõesdiferentespelasquaisumartigoérejeitado,amaioriadelasevitáveis.

•  Algumasdasprincipaisrazõespararejeiçãosãolistadasaseguir.

•  Cadaumadelaséigualmenteimportante,poisosrevisorestendemafocaremquestõesdiferentesdependendodassuaspreocupaçõeseconhecimentosindividuais.

Desenhoexperimentalpobree/ouinvestigaçãoinadequada

•  Amostrascomtamanhosinadequados;•  Amostragemenviesada;•  Umoumaisconceitosambíguos;•  Incorreçõescientíficas.

Artigonãoadequadoàrevista

•  Ofocodoartigonãoestádeacordocomoescopodarevista;

•  Asguidelinesdarevistanãoforamseguidas.•  Istopodeserfacilmenteevitávelconsultando-sepreviamentealgunsartigosdarevistaeasinstruçõesparaosautores.

Errosgramaticaisedeestilonotextoeminglês

•  Umaescritapobrenãoimplicaemumarejeiçãoimediatadoartigo,maspodeinfluenciarbastantenaimpressãogeraldosrevisoresedoeditorsobreele.

•  Estudosmostramqueartigosbemescritostêmmaischancesdeseraceitos.

Definiçãoinsuficientedoproblema

•  Éimportantedefinirclaramenteecircunscreverdeformaapropriadaaquestãoestudada.

Métodospoucodetalhados•  Osdetalhesdescritossãoinsuficientesparaserepetirosresultados.

•  Odesenhodoestudo,osequipamentosusadoseosprocedimentosdevemserexplicadosdeformaclara.

•  Emalgunscasos,émelhorpôrinformaçãoemexcessonaseçãodemétodosdoquepôrmuitopouca;ainformaçãoconsideradadesnecessáriasemprepodeserretiradaantesdapublicação.

Interpretaçãoexageradadovalordosresultados

•  Umaabordagemclarae“honesta”dosresultadosaumentaaprobabilidadedoartigoseraceito.

•  Identifiquepossíveisviésesevariáveisconfusas,tantonafasededesenhodoestudocomonadeinterpretaçãodosresultados.

•  Sejaconcisoaodescreverosresultadosexperimentais.

Análiseestatísticainapropriadaouincompleta

•  Usetestesestatísticosapropriadosenãotorneaanáliseestatísticamuitocomplicada.

•  Quantifiqueeapresenteosresultadoscomindicadoresapropriadosdeerrodemedidaedeincerteza(intervalosdeconfiança,porexemplo).

Apresentaçãoinsatisfatóriaouconfusadedadosemtabelasefiguras

•  Osestilosdastabelasefigurasnãoestãoconformeasguidelinesdarevista.

•  Astabelasefigurasestãosobrecarregadasdenúmerosesímbolos.

•  Façaastabelasefigurasfáceisdeseremlidas.

Conclusõesnãosustentadaspelosdados

•  Certifique-sedequeasconclusõesnãosãoexageradas,quetêmapoionosresultadoserespondemàsquestõespostasnoestudo.

•  Tenhacertezadeterdiscutidoexplicaçõesalternativasenãofaçadasconclusõesapenasumarepetiçãodosresultadoscompalavrasdiferentes.

Revisãodaliteraturaincompleta,imprecisaedesatualizada

•  Certifique-sedeterfeitoumarevisãocompletadaliteraturaelisteapenasasreferênciasrelevantesparaoestudo.

•  Osrevisoresdoartigoserãoespecialistasnoassuntoeestarãocientesdetodasaspesquisasrelevantesjáfeitas.

Faltadeatençãoàssugestõesdosrevisores

•  Acatarassugestõesdosrevisoresnahoraderevisaroartigoiráquasesempreresultaremumartigomelhor.

•  Seoeditorindicarumapredisposiçãoparaavaliarumaversãorevisadadoartigo,istosignificaqueeleépublicávelseaspreocupaçõesdosrevisoresforematendidasdeformasatisfatória.

Checklistparasubmissão

•  Éimportanteprepararoartigodeformaapropriada,oqueimplicaseguirasguidelinesdarevista.Ousodeumchecklistajudaagarantiraaceitaçãodoartigopelarevista.

•  Dá-seaseguirumchecklistgenéricoparagarantirqueoartigoestejadeacordocomasguidelinesdamaioriadasrevistas.

Carta de encaminhamento [] Determine se uma carta de encaminhamento (cover letter) é necessária. [] Dirija-se ao editor correto da revista de acordo com o assunto do artigo. [] Use o endereço correto. [] Revise o que é necessário na carta de encaminhamento.

Modelo de carta de encaminhamento Dr. James S. Goodwine Editor-In-Chief, Journal of Random Issues The Chance Institute - NAC 10010 East Vinery Road Barley Bush, CA 92066 USA Dear Dr. Goodwine, Enclosed please find three copies of the article A realistic computer simulation of the effect of the mackerel tail on sea waves to be considered for publication in Journal of Random Issues. A set of original figures is also enclosed. This article is based on an ongoing research aimed at a PhD degree, which the graduate student José Sardinha is doing at the Department of Physics of the University of Caixa-Prego, Brazil, under my supervision. Provided that the article is found to be satisfactory, we would be pleased to provide the source code (in Microsoft Word for Windows XP, version 2003) to expedite its publication. Please send all the correspondence relative to this article to me. Yours Sincerely Dr. Jonas Baleia

Geral [] Determine que tipo de artigo você está submetendo (letter, review, original scientific paper, etc). [] Use os tipos corretos de fonte e de tamanho para as letras. [] Ajuste os espaçamentos entre linhas (simples ou duplos). [] Verifique os formatos dos títulos das seções. [] Ponha as seções na ordem correta. [] Verifique os limites de números de palavras. [] Use numeração de linhas, caso pedido pela revista. [] Use números nas páginas, caso pedido pela revista. [] Ajuste os tamanhos das margens. [] Confirme se a nomenclatura está correta. [] Verifique a grafia das palavras. [] Determine se as seções de resultados e de discussão estão separadas ou juntas em uma única seção.

Página do Título [] Verifique se o título está com o tamanho permitido. [] Determine se é necessária a inclusão de um título curto (running title). [] Verifique se são necessárias palavras-chave (keywords). [] Confirme se a revista pede uma lista de abreviações. [] Certifique-se de que todos os autores estão incluídos. [] Verifique se os nomes e endereços dos autores estão com os formatos corretos. [] Inclua todas as informações sobre o autor correspondente. Abstract [] Confirme se está dentro do limite de palavras. [] Determine se a revista pede um abstract estruturado ou não estruturado. Referências [] Confirme que todas as citações feitas no texto estão no formato correto. [] Verifique se todas as referências citadas no texto estão incluídas na lista de referências. [] Confirme que todas as referências na lista de referências estão citadas no texto. [] Determine se as referências estão formatadas corretamente. [] Verifique se os detalhes das referências (volume, ano, páginas, nomes dos autores, etc) estão corretos.

Tabelas e Figuras [] Verifique se as menções às figuras e tabelas feitas no texto estão formatadas corretamente. [] Determine se as tabelas e figuras estão posicionadas corretamente. [] Verifique se as fontes e tamanhos de fontes das palavras e símbolos nas tabelas e figuras estão corretas. [] Confirme se a numeração das tabelas e figuras está no formato correto (números romanos ou arábicos). [] Verifique se os tamanhos das figuras e tabelas estão corretos. [] Verifique se os arquivos estão nos formatos corretos (pdf, jpeg, gif, etc). [] Determine o tipo de lista para os títulos das tabelas e as legendas das figuras. [] Certifique que todas as tabelas e figuras estão mencionadas no texto. [] Determine se a revista permite linhas verticais nas tabelas. Miscelânea [] Determine se será preciso uma declaração sobre conflito de interesses. [] Verifique se é necessária uma menção às fontes de financiamento. [] Para artigos na área médica: inclua todas as declarações de aprovações pelos comitês de ética e pelos pacientes.

PromovendosuaPublicação•  Apublicaçãodeumartigoemumarevistacientíficanãogarantequeeleserálidooucitado.

•  Aquantidadedeartigospublicadosdiariamenteétãograndequeamaioriadoscientistasnãotemtempoparalertudo.

•  Comoelessaberãoquevocêacaboudedarumacontribuiçãoimportanteàáreasenãoestiveremprocurandoespecificamenteporseuartigo?

•  Apósapublicaçãodeumartigo,hámedidasaseremtomadasparagarantirqueoartigosejadistribuídoeatraiaaatençãodepessoasrelevantesdentrodassuasredesdecontatosacadêmicos.

•  Umamaneiradedivulgaroartigoémandarcópiasdeleparaaspessoas.

Enviecópiasdoartigopara:•  Autorescitadosnoartigo;•  Pesquisadoresecientistasquepublicaramnomesmoassuntoouqueestejamtrabalhandonamesmaárea;

•  Pessoas/organizaçõesqueapoiaramasuapesquisa;

•  Bibliotecadasuainstituição;•  Gruposdediscussãoonlinevoltadosparaagrandeáreadapesquisa.

•  Éútilterumalistadessaspessoaseiracrescentandonovosnomesàmedidaqueforexpandindosuarededecontatos.

•  Amaioriadasrevistasofereceumserviçodereprints(enviamumnúmerosolicitadodecópiasextrasdoartigomediantepagamento);sevocêtiverumalistadepessoas,saberáquantosreprintspedir.

•  Seformandarcópiasparaadministradoresepessoaldepolíticacientífica,deveprepararumacartadeencaminhamentoresumindooartigoemlinguagemnãotécnicaexplicandosuaimportância.

•  Éimportanteterumapáginawebemsuainstituiçãoouporcontaprópriaondecolocarseusartigospublicadosdeumaformabaixável.

•  Estaéumamaneiradeeconomizarnocustodosreprints.