解决方案如下：
- Download Genbank from NCBI;
- Convert Genbank to GFF3 with BioPerl;
- Modify the GFF3 file;
- Download Fasta from NCBI or use Genbank conversion;
- Use the modified GFF3 file in rsem-prepare-reference;
主要代码如下：
bp_genbank2gff3 --GFF_VERSION 3 file.gb
awk '$3!~/pseudo/' file.gb.gff > file.rsem.gff
# or more restrictly with only protein coding genes
awk '$3~/^(gene|mRNA|exon|polypeptide)$/' file.gb.gff > file.cds.gff
# remove the version number in fasta accession if no transcript is extract
rsem-prepare-reference file.fasta --gff3 file.rsem.gff reference_name
这里主要解决了以下错误：
Line x does not have attribute Parent
由于rsem遵循gene->transcript->exon的结构，并利用exon来作为RNA的最后形式（经过剪切和后处理的RNA），要求exon都有(且仅有）一个Parent。而GFF3文件中pseudogene和pseudogenic_exon是没有Parent的。
以下为rsem-gff3-to-gtf line 27-32。细菌基因组的GFF3文件在gene->transcript->exon中的映射关系如下：gene = type_gene, mRNA(tRNA, etc) = type_transcript, CDS = exon.
type_gene = ["gene", "snRNA_gene", "transposable_element_gene", "ncRNA_gene", "telomerase_RNA_gene",
"rRNA_gene", "tRNA_gene", "snoRNA_gene", "mt_gene", "miRNA_gene", "lincRNA_gene", "RNA", "VD_gene_segment"]
type_transcript = ["transcript", "primary_transcript", "mRNA", "ncRNA", "tRNA", "rRNA", "snRNA", "snoRNA", "miRNA",
"pseudogenic_transcript", "lincRNA", "NMD_transcript_variant", "aberrant_processed_transcript",
"nc_primary_transcript", "processed_pseudogene", "mRNA_TE_gene"]
type_exon = ["exon", "CDS", "five_prime_UTR", "three_prime_UTR", "UTR", "noncoding_exon", "pseudogenic_exon"]
