do_prediction=function(iso,gene.marker,reftranscriptome,grid,t)
{

	glist=gene.marker[is.na(iso)==0]
	iso1=iso[is.na(iso)==0]
	a=grid$a
	b=grid$b
	c=grid$c
	
	
	
	lk=NULL
	k=1
	
	for(i in 1:length(a))
	{
	    y=iso1-a[i]*gam[glist,c[i]]-(1-a[i])*reftranscriptome[glist,b[i]]
	    lk[k] = -sum(dnorm(y[is.na(y)==0],mean(y[is.na(y)==0]), 0.8,log=T))
	    k=k+1
	}

	inde=order(lk)[1]
	aa=a[inde]
	s=aa-0.1
	e=aa+0.1
	if(aa ==0)
	{
	    s=0
	}

	lk=NULL
	k=1
	aaa=seq(s,e,by=0.001)
	for(i in aaa)
	{
	    y=iso1-i*gam[glist,c[inde]]-(1-i)*reftranscriptome[glist,b[inde]]
	    lk[k] = -sum(dnorm(y[is.na(y)==0],mean(y[is.na(y)==0]), 0.8,log=T))
	    k=k+1
	}

	r=c(t[b[inde]]*2,aaa[order(lk)[1]])

	return(r)
}

predict_age_gam=function(d,outfile)
{
	refdd2=read.table("DDd2ref.txt",header=T,sep="\t",na.string="NA")
	gam=read.table("GAMref.txt",header=T,na.string="NA",sep="\t")
	gene.marker=intersect(intersect(rownames(d),rownames(gam)),rownames(refdd2))
	
	t=(0:180)/10
	refdd2.100tps=t(apply(refdd2[gene.marker,],1,function(x) predict(smooth.spline(1:24,x,df=5),t)$y))

	grid_a0=seq(0,0.8,by=0.1)
	grid_t=21:131
	grid_gam=5:12

	a=NULL
	b=NULL
	c=NULL
	for(i in grid_a0)
	{
		for(j in grid_t)
		{
			for(m in grid_gam)
			{
				 a=c(a,i)
				 b=c(b,j)
				 c=c(c,m)
			}
		}
	}
	grid=cbind(a,b,c)

	pred=apply(d[gene.marker,],2,function(x) do_prediction(x,gene.marker,refdd2.100tps,grid,t))


	rr=t(pred)
	rownames(rr)=colnames(pred)
	colnames(rr)=c("esti.hpi","esti.gam.prop")
	write.table(rr,file="Mixture_model_hpigam_prediction_GAM.txt",row.names=T,col.names=T,sep="\t",quote=F)
}
--------------------------------------------------------------------
library(stats4)
isolate=read.table("isolate_transcriptome.txt",header=T,sep="\t",na.string="NA")
predict_age_gam(d,"outfile.txt")
#-----------------------------------------------------------------------