EU-kalat

EU-kalat is a study, where concentrations of PCDD/Fs, PCBs, PBDEs and heavy metals have been measured from fish

Question

The scope of EU-kalat study was to measure concentrations of persistent organic pollutants (POPs) including dioxin (PCDD/F), PCB and BDE in fish from Baltic sea and Finnish inland lakes and rivers. .

Answer

Dioxin concentrations in Baltic herring.

The original sample results can be acquired from Opasnet base. The study showed that levels of PCDD/Fs and PCBs depends especially on the fish species. Highest levels were on salmon and large sized herring. Levels of PCDD/Fs exceeded maximum level of 4 pg TEQ/g fw multiple times. Levels of PCDD/Fs were correlated positively with age of the fish.

Mean congener concentrations as WHO2005-TEQ in Baltic herring can be printed out with this link or by running the codel below.

+ Show code- Hide code # This code is Op_en on page [[EU-kalat]]. library(OpasnetUtils) library(ggplot2) eu <- Ovariable("eu", ddata = "Op_en3104", subset = "POPs") eu <- EvalOutput(eu) #levels(eu$POP) eu <- eu[eu$Fish_species == "Baltic herring" , ] tef <- Ovariable("tef", ddata = "Op_en4017", subset = "TEF values") tef <- EvalOutput(tef) colnames(eu@output)[colnames(eu@output) == "POP"] <- "Congener" levels(eu$Congener) <- gsub("HCDD", "HxCDD", levels(eu$Congener)) levels(eu$Congener) <- gsub("HCDF", "HxCDF", levels(eu$Congener)) levels(eu$Congener) <- gsub("CoPCB", "PCB", levels(eu$Congener)) #levels(eu$Congener) euteq <- eu * tef #head(euteq@output) temp <- aggregate(euteq@output["Result"], by = c(euteq@output["Congener"], euteq@output["Catch_location"]), FUN = mean) temp2 <- temp temp2$Result <- temp2$Result / sum(temp2$Result) cat("Congener TEQ profile of Baltic herring.\n") oprint(temp2) ggplot(temp, aes(x = Congener, y = Result, fill = Congener)) + geom_bar(stat = "identity") + facet_wrap(~ Catch_location) + labs(x = "Congener", y = "TEQ concentration (pg/g fat)") + theme_gray(base_size = 18) + coord_flip() temp2 <- temp[temp$Congener %in% levels(temp$Congener)[1:17],] ggplot(temp2, aes(x = Congener, y = Result, fill = Congener)) + geom_bar(stat = "identity") + facet_wrap(~ Catch_location) + labs( title="Dioxin concentrations in Baltic herring", x = "Congener", y = "TEQ concentration (pg/g fat)" ) + theme_gray(base_size = 18) + coord_flip()

Rationale

Data

Data was collected between 2009-2010. The study contains years, tissue type, fish species, and fat content for each concentration measurement. Number of observations is 285.

There is a new study EU-kalat 3, which will produce results in 2016.

Calculations

Preprocess

• Preprocess model 22.2.2017 [4]
  • Objects used in Benefit-risk assessment of Baltic herring and salmon intake
• Model run 25.1.2017 [5]
• Model run 22.5.2017 with new ovariables euRaw, euAll, euMain, and euRatio [6]
• Model run 23.5.2017 with adjusted ovariables euRaw, eu, euRatio [7]
• Model run 11.10.2017: Small herring and Large herring added as new species [8]
• Model rerun 15.11.2017 because the previous stored run was lost in update [9]
• Model run 21.3.2018: Small and large herring replaced by actual fish length [10]

+ Show code- Hide code # This is code Op_en3104/preprocess on page [[EU-kalat]] library(OpasnetUtils) library(ggplot2) library(reshape2) # Divide herring into small and large (as new "species") euRaw <- Ovariable("euRaw", ddata = "Op_en3104", subset = "POPs") #euRaw <- EvalOutput(euRaw) #temp <- euRaw[euRaw$Fish_species == "Baltic herring" , ] #temp$Fish_species <- as.factor( # ifelse( # as.numeric(as.character(temp$Length_mean_mm)) >= 170, # "Large herring", # "Small herring" # ) #) #euRaw <- combine(euRaw, temp, name = "euRaw") # levels(TEF$Group) #[1] "Chlorinated dibenzo-p-dioxins" "Chlorinated dibenzofurans" #[3] "Mono-ortho–substituted PCBs" "Non-ortho–substituted PCBs" indexguide <- Ovariable(data=data.frame( # This is needed to make these indices marginals in eu (which is in long format). Length = 1, Year = 1, Weight = 1, Large = 1, Result = 1 )) eu <- Ovariable( "eu", dependencies = data.frame( Name=c("euRaw", "TEF", "indexguide"), Ident=c(NA,"Op_en4017/initiate", NA) ), formula = function(...) { euRaw$Length<-as.numeric(as.character(euRaw$Length_mean_mm)) euRaw$Year <- as.numeric(substr(euRaw$Catch_date, nchar(as.character(euRaw$Catch_date))-3,100)) euRaw@marginal[colnames(euRaw@output) %in% c("Length","Year")] <- TRUE # euRaw$Weight<-as.numeric(as.character(euRaw$Weight_mean_g)) # euRaw$Large <- euRaw$Length >= 170 eu <- euRaw[,c(1:4, 10, 20, 21, 18)] # See below + Length, Year, Result colnames(eu@output)[1:5] <- c("THLcode", "Matrix", "Compound", "Fish", "N") temp <- oapply(eu * TEF, cols = "Compound", FUN = "sum") colnames(temp@output)[colnames(temp@output)=="Group"] <- "Compound" eu <- combine(eu, temp) eu$Compound <- factor( # Compound levels are ordered based on the data table on [[TEF]] eu$Compound, levels = unique(c(levels(TEF$Compound), levels(eu$Compound))) ) eu$Compound <- eu$Compound[,drop=TRUE] return(eu) } ) euRatio <- Ovariable( "euRatio", dependencies = data.frame(Name=c("eu")), formula = function(...) { euRatio <- eu[ eu$Compound == "2378TCDD" & eu$Matrix == "Muscle" & result(eu) != 0 , ] # Zeros cannot be used in ratio estimates euRatio$Compound <- NULL euRatio <- log10(eu / euRatio)@output euRatio <- euRatio[!euRatio$Compound %in% c("2378TCDD", "2378-TCDD", "TCDD") , ] return(euRatio) } ) # Analysis: a few rows disappear here, as shown by numbers per fish species. Why? # aggregate(eut@output, by = eut@output["Fish"], FUN = length)[[2]] # [1] 1181 141 131 55 158 51 96 30 36 40 102 49 61 180 eu # [1] 1173 141 131 55 158 51 96 27 36 40 102 49 53 180 eu/eut # There are samples where TCDD concenctration is zero. #eu@data[eu@data$POP == "2378TCDD" & eu@data$euResult == 0 , c(1,4)][[1]] #eu@data[eu@data[[1]] %in% c("09K0585", "09K0698", "09K0740", "09K0748") , c(1,3,4,18)] # Some rows disappear because there is no TCDD measurement to compare with: #8 Baltic herrings, 8 sprats, 3 rainbow trouts. # Conclusion: this is ok. Total 2292 rows. ################## Data for the main congeners and species only #> unique(eu$Congener) #[1] 2378TCDD 12378PeCDD 123478HCDD 123678HCDD 123789HCDD 1234678HpCDD #[7] OCDD 2378TCDF 12378PeCDF 23478PeCDF 123478HCDF 123678HCDF #[13] 123789HCDF 234678HCDF 1234678HpCDF 1234789HpCDF OCDF CoPCB77 ... # Remove the four PCDDFs with too little data (>70% BDL) and all non-PCDDF # aggregate(eu@data$euResult, by = eu@data["POP"], FUN = function(x) mean(x == 0)) #[1] Baltic herring Sprat Salmon Sea trout Vendace #[6] Roach Perch Pike Pike-perch Burbot #[11] Whitefish Flounder Bream River lamprey Cod #[16] Trout Rainbow trout Arctic char Small herring Large herring indices <- list( Compound.TEQ2 = c("PCDDF", "PCB"), Compound.PCDDF14 = as.character(unique(euRaw$POP)[c(1:12, 14, 15)]), # 7 OCDD should be removed Fish.Fish16 = as.character(unique(euRaw$Fish_species)[c(1:4, 6:14, 17, 19, 20)]) ) #> indices #$Compound.TEQ2 #[1] "PCDDF" "PCB" # #$Compound.PCDDF14 # [1] "2378TCDD" "12378PeCDD" "123478HCDD" "123678HCDD" "123789HCDD" "1234678HpCDD" # [7] "OCDD" "2378TCDF" "12378PeCDF" "23478PeCDF" "123478HCDF" "123678HCDF" #[13] "234678HCDF" "1234678HpCDF" # #$Fish.Fish16 # [1] "Baltic herring" "Sprat" "Salmon" "Sea trout" "Roach" # [6] "Perch" "Pike" "Pike-perch" "Burbot" "Whitefish" #[11] "Flounder" "Bream" "River lamprey" "Rainbow trout" "Small herring" #[16] "Large herring" #ggplot(euRaw@output, aes(x=Length, y=Weight))+geom_point()+ # facet_wrap(~Fish_species, scales="free") #oprint(summary( # euRaw[euRaw$Fish_species %in% c("Baltic herring","Salmon") & euRaw$POP == "2378TCDD" , ], # marginals=c("Catch_location","Large","Fish_species"), # fun = "length" #)) objects.store(euRaw, eu, euRatio, indices, indexguide) cat("Ovariables euRaw, eu, euRatio, and indexguide and list indices stored.\n")

Bayes model for dioxin concentrations

• Model run 28.2.2017 [11]
• Model run 28.2.2017 with corrected survey model [12]
• Model run 28.2.2017 with Mu estimates [13]
• Model run 1.3.2017 [14]
• Model run 23.4.2017 [15] produces list conc.param and ovariable concentration
• Model run 24.4.2017 [16]
• Model run 19.5.2017 without ovariable concentration [17] ⇤--#: . The model does not mix well, so the results should not be used for final results. --Jouni (talk) 19:37, 19 May 2017 (UTC) (type: truth; paradigms: science: attack)

----#: . Maybe we should just estimate TEQs until the problem is fixed. --Jouni (talk) 19:37, 19 May 2017 (UTC) (type: truth; paradigms: science: comment)

• Model run 22.5.2017 with TEQdx and TEQpcb as the only Compounds [18]
• Model run 23.5.2017 debugged [19] [20] [21]
• Model run 24.5.2017 TEQdx, TECpcb -> PCDDF, PCB [22]
• Model run 11.10.2017 with small and large herring [23] (removed in update)
• Model run 12.3.2018: bugs fixed with data used in Bayes. In addition, redundant fish species removed and Omega assumed to be the same for herring and salmon. [24]
• Model run 22.3.2018 [25] Model does not mix well. Thinning gives little help?

+ Show code- Hide code # This is code Op_en3104/bayes on page [[EU-kalat]] library(OpasnetUtils) library(reshape2) library(rjags) # JAGS library(ggplot2) library(MASS) # mvrnorm library(car) # scatterplotMatrix objects.latest("Op_en3104", code_name = "preprocess") # [[EU-kalat]] eu, euRatio, indices eu2 <- eu <- EvalOutput(eu) conl <- indices$Compound.TEQ2 # fisl <- indices$Fish.Fish16 # fisl <- c("Baltic herring","Large herring","Salmon","Small herring") fisl <- c("Baltic herring","Salmon") C <- length(conl) Fi <- length(fisl) N <- 1000 conl fisl replaces <- list( c("Chlorinated dibenzo-p-dioxins", "PCDDF"), c("Chlorinated dibenzofurans", "PCDDF"), c("Mono-ortho-substituted PCBs", "PCB"), c("Non-ortho-substituted PCBs", "PCB") ) for(i in 1:length(replaces)) { levels(eu2$Compound)[levels(eu2$Compound)==replaces[[i]][1]] <- replaces[[i]][2] } eu2@marginal[colnames(eu2@output) %in% c("Length","Year")] <- TRUE # Indexguide should take care of this but it doesn't! eu2 <- unkeep(eu2, prevresults = TRUE, sources = TRUE) eu2 <- oapply(eu2, cols = "TEFversion", FUN = "sum") # Sums up dioxin+furan and non+monoortho. This goes wrong if > 1 TEFversion. # Hierarchical Bayes model. # PCDD/F concentrations in fish. # It uses the TEQ sum of PCDD/F (PCDDF) as the total concentration # of dioxin and PCB respectively for PCB in fish. # PCDDF depends on age of fish, fish species and catchment area, # but we only have species now so other variables are omitted. # cong depends on fish species. eu3 <- eu2[eu2$Compound %in% conl & eu2$Fish %in% fisl & eu2$Matrix == "Muscle" , ] eu3 <- reshape( # Gives warning that more than one value and first taken. Why? eu3@output, v.names = "euResult", idvar = c("THLcode", "Fish"), timevar = "Compound", drop = c("Matrix"), direction = "wide" ) oprint(head(eu3)) #> colnames(eu3) #[1] "THLcode" "Fish" "N" "Length" "Year" #[6] "euResult.PCDDF" "euResult.PCB" cong <- data.matrix(eu3[6:ncol(eu3)]) if(FALSE){ # Find the level of quantification for dinterval function LOQ <- unlist(lapply(eu3[6:ncol(eu3)], FUN = function(x) min(x[x!=0]))) # With TEQ, there are no zeros. So this is useful only if there are congener-specific results. names(LOQ) <- conl cong <- sapply( 1:length(LOQ), FUN = function(x) ifelse(cong[,x]==0, 0.5*LOQ[x], cong[,x]) ) } # This version of the model looks only at Baltic herring, Large herring, small herring and salmon. # It assumes that all fish groups have the same Omega but mu varies. mod <- textConnection( " model{ for(i in 1:S) { # S = fish sample # below.LOQ[i,j] ~ dinterval(-cong[i,j], -LOQ[j]) cong[i,1:C] ~ dmnorm(muind[i,], Omega[,]) muind[i,1:C] <- mu[fis[i],1:C] + lenp[fis[i]]*length[i] # + timep*year[i] } # Priors for parameters # timep ~ dnorm(0, 0.0001) # Time trend. We don't want to use it yet for(i in 1:Fi) { # Fi = fish species lenp[i] ~ dnorm(0,0.0001) # length parameter pred[i,1:C] ~ dmnorm(mu[i,1:C]+lenp[i]*lenpred[i], Omega[,]) # Model prediction. Not used: +timep*timepred for(j in 1:C) { mu[i,j] ~ dnorm(0, 0.0001) # mu1[j], tau1[j]) # Congener-specific mean for fishes } } Omega[1:C,1:C] ~ dwish(Omega0[1:C,1:C],S) } ") jags <- jags.model( mod, data = list( S = nrow(eu3), C = C, Fi = Fi, cong = log(cong), length = eu3$Length, # year = eu3$Year, fis = match(eu3$Fish, fisl), lenpred = c(170, 860), # timepred = 2009, Omega0 = diag(C)/100000 ), n.chains = 4, n.adapt = 100 ) update(jags, 1000) samps.j <- jags.samples( jags, c( 'mu', # mean by fish and compound 'Omega', # precision matrix by compound 'lenp',# parameters for length 'timep', # parameter for Year 'pred' # predicted concentration for year 2009 and 17 cm herring, 80 cm salmon ), # thin=1000, N ) dimnames(samps.j$mu) <- list(Fish = fisl, Compound = conl, Iter = 1:N, Chain = 1:4) dimnames(samps.j$lenp) <- list(Fish = fisl, Iter = 1:N, Chain = 1:4) dimnames(samps.j$pred) <- list(Fish = fisl, Compound = conl, Iter = 1:N, Chain = 1:4) dimnames(samps.j$Omega) <- list(Compound = conl, Compound2 = conl, Iter=1:N, Chain=1:4) ##### conc.param contains expected values of the distribution parameters from the model conc.param <- list( Omega = apply(samps.j$Omega[,,,1], MARGIN = 1:2, FUN = mean), lenp.mean = apply(samps.j$lenp[,,1], MARGIN = 1, FUN = mean), lenp.sd = apply(samps.j$lenp[,,1], MARGIN = 1, FUN = sd), mu = apply(samps.j$mu[,,,1], MARGIN = 1:2, FUN = mean) # timep.mean = apply(samps.j$lenp[,,1], MARGIN = 1, FUN = mean), # timep.sd = apply(samps.j$lenp[,,1], MARGIN = 1, FUN = sd) ) objects.store(conc.param, samps.j) cat("Lists conc.params and samps.j stored.\n") ######################3 cat("Descriptive statistics:\n") # Leave only the main fish species and congeners and remove others #oprint(summary( # eu2[eu2$Compound %in% indices$Compound.PCDDF14 & eu$Fish %in% fisl , ], # marginals = c("Fish", "Compound"), # Matrix is always 'Muscle' # function_names = c("mean", "sd") #)) ggplot(eu2@output[eu2$Fish %in% fisl & eu2$Compound %in% conl,], aes(x = euResult, colour=Compound))+stat_ecdf()+ facet_wrap( ~ Fish)+scale_x_log10() ggplot(melt(exp(samps.j$pred[,,,1])), aes(x=value, colour=Compound))+stat_ecdf()+ facet_wrap( ~ Fish)+scale_x_log10() scatterplotMatrix(t(samps.j$pred[1,,,1]), main = "Predictions for all compounds for Baltic herring") scatterplotMatrix(t(samps.j$pred[,1,,1]), main = "Predictions for all fish species for PCDDF") coda.j <- coda.samples( jags, c('mu', 'pred','lenp'), # thin=1000, N ) plot(coda.j) coda.o <- coda.samples( jags, c('Omega'), N ) plot(coda.o)

Initiate conc_pcddf

• Model run 19.5.2017 [26]
• Model run 23.5.2017 with bugs fixed [27]
• Model run 12.10.2017: TEQ calculation added [28]
• Model rerun 15.11.2017 because the previous stored run was lost in update [29]
• 12.3.2018 adjusted to match the same Omega for all fish species [30]

+ Show code- Hide code # This is code Op_en3104/initiate on page [[EU-kalat]] library(OpasnetUtils) conc_pcddf <- Ovariable( "conc_pcddf", dependencies = data.frame(Name = "conc.param", Ident = "Op_en3104/bayes"), formula = function(...) { require(MASS) require(reshape2) jsp <- lapply(1:length(conc.param$mu[, 1]), FUN = function(x) { temp <- exp(mvrnorm( openv$N, conc.param$mu[x, ], solve(conc.param$Omega[ , ]) )) dimnames(temp) <- c(list(Iter = 1:openv$N), dimnames(conc.param$mu)[2]) return(temp) }) names(jsp) <- dimnames(conc.param$mu)[[1]] jsp <- melt(jsp, value.name = "Result") colnames(jsp)[colnames(jsp) == "L1"] <- "Fish" jsp <- Ovariable( output = jsp, marginal = colnames(jsp) != "Result" ) d <- oapply(jsp, cols = "Compound", FUN = sum) d$Compound <- "TEQ" return(combine(jsp,d)) } ) objects.store(conc_pcddf) cat("Ovariable conc_pcddf stored.\n")

⇤--#: . These codes should be coherent with POPs in Baltic herring. --Jouni (talk) 12:14, 7 June 2017 (UTC) (type: truth; paradigms: science: attack)