I have 10 sample with two condition (case/controll).  I aligned using STAR 2.25a . On the  left image after aligned I count using htseq-count and then I select only counts are from protein-coding and I use for generate pca plot. In the second case (right image) I use rsem with STAR and I use to import the counts using tximport. In the first case I use ensemble 74 in the second case hg38/84. Both pca are different  What could be wrong? In the first case seem the date of the run do not affect pca (Using htseqcount) in the second case PC1 are able to divide one batch .

Here the script for prepare the counts that I use (only protein coding)

ret<-data.frame(rownames(dt))
ret$ensembl <- sapply(strsplit(rownames(dt),split="nn+"),"[",1)
#use biomart

#ensembl = useMart( "ensembl", dataset = "hsapiens_gene_ensembl" )
ensembl = useMart( host="dec2013.archive.ensembl.org",biomart="ENSEMBL_MART_ENSEMBL", dataset = "hsapiens_gene_ensembl" )

genemap <- getBM( attributes = c("ensembl_gene_id", "hgnc_symbol","family","gene_biotype","family_description"),
                  filters = "ensembl_gene_id",
                  values = list(ret$ensembl,"protein_coding"),
                  mart = ensembl )
idx <- match( ret$ensembl, genemap$ensembl_gene_id )

ret$hgnc_symbol <- genemap$hgnc_symbol[ idx ]
ret$entrez <- genemap$entrezgene[ idx ]
ret$hgnc_symbol <- genemap$hgnc_symbol[ idx ]
ret$family <- genemap$family[ idx ]
ret$biotype <- genemap$gene_biotype[ idx ]
ret$family_description <- genemap$family_description[ idx ]

#prepare  coding

matrice1<-ret[ret$biotype=="protein_coding",]
head(matrice1$ensembl)
countRaw<-as.data.frame(counts(dds,normalized=FALSE))
countRawCoding <-  countRaw[matrice1$ensembl,]

and for the plot pca

http://imgur.com/6CKCUPq