Output: intron/UTR/CDS in bed12 format (which can be further converted into bed6 via "bedtools bed12tobed6" command)
# Introns (if any) of gene annotation in BED12 format
cat annotation.bed | awk '{OFS="\t";split($11,a,","); split($12,b,","); A=""; B=""; for(i=1;i<length(a)-1;i++) {A=A""(b[i+1]-b[i]-a[i])",";B=B""(b[i]+a[i]-(b[1]+a[1]))",";} if($10>1) print $1,$2+a[1], $3-a[length(a)-1], $4,$5,$6,$2+a[1], $3-a[length(a)-1],$9,$10-1,A,B;}'
# 5´ UTR (if any) of gene annotation in BED12 format
## Update: the code crossed above did not consider the special case that start codon is spliced (e.g. in intron within in the start codon).
## Update (2017-Jan-28, see comment below)
awk '{OFS="\t";split($11,blockSizes,","); split($12,blockStarts,","); blockCount=$10;A=""; B=""; if($7==$8) next;N=0;if($6=="+" && $2<$7) {start=$2;end=$7; for(i=1;i<=blockCount;i++) if(($2+blockStarts[i]+blockSizes[i])<=$7) {A=A""blockSizes[i]",";B=B""blockStarts[i]","; end=($2+blockStarts[i]+blockSizes[i]); N++;} else { if(($2+blockStarts[i])<$7) {A=A""($7-$2-blockStarts[i])",";B=B""blockStarts[i]","; N++; end=$7;} break; } print $1,start,end,$4,$5,$6,start,end,$9,N,A,B;} if($6=="-" && $8<$3) {start=$8;end=$3; for(i=1;i<=blockCount;i++) if(($2+blockStarts[i])>=$8) {if(start==0) {A=blockSizes[i];B=0; start=$2+blockStarts[i];} else {A=A","blockSizes[i];B=B","($2+blockStarts[i]-start);} N++;} else { if(($2+blockStarts[i]+blockSizes[i])>$8) { A=($2+blockStarts[i]+blockSizes[i]-$8);B=0; N++; start=$8;} if(($2+blockStarts[i]+blockSizes[i])==$8) start=0;} print $1,start,end,$4,$5,$6,start,end,$9,N,A,B;}}'
# CDS (if any) of gene annotation in BED12 format
grep "protein_coding\.protein_coding" annotation.bed | awk '{OFS="\t";split($11,a,","); split($12,b,","); A=""; B=""; if($7==$8) next; j=0; for(i=1;i<length(a);i++) if(($2+b[i]+a[i])>$7 && ($2+b[i])<$8) {j++; start=$2+b[i]-$7; size=a[i]; if(($2+b[i])<=$7) {start=0;size=size-($7-($2+b[i]));} if(($2+a[i]+b[i])>=$8) {size=size-($2+a[i]+b[i]-$8);} A=A""size",";B=B""start",";} print $1,$7,$8,$4,$5,$6,$7,$8,$9,j,A,B;}'
### Note1: The thickStart and thickEnd in the bed12 format don't always indicate CDS. "When there is no thick part, thickStart and thickEnd are usually set to the chromStart position." So, we cannot use bed12 to infer CDS (or coding exons), esp. for lincRNA. The correct way is to grep all CDS lines from the GTF file, or directly using UCSC Table Browser to download coding exons.
### Note2: The "Polymorphic pseudogene" in GENCODE can also have a protein-coding transcript, even though the gene itself is classified as a pseudogene (YES, polymorphic pseudogenes are also pseudogenes, they "are coding gene that are pseudogenic due to the presence of a polymorphic premature stop codon in the reference genome" (http://www.genomebiology.com/2012/13/9/R51). For more details, see https://gencodegenes.wordpress.com/toolbox/.
# 3´ UTR (if any) of gene annotation in BED12 format
Update: Thanks to Heather's comment below, when the thickEnd (end of CDS) is at the end of an exon (or thickStart at the first nt of an exon), it will generate an invalid BED12 format. It's fixed now. (2017-Jan-28)
BTW, the above codes are integrated into a neat script in Github: https://github.com/sterding/RNAseq/blob/master/bin/bed12toAnnotation.awk
Also, I fixed some bugs in gtf2bed (originally by Erik) and host here:
https://github.com/sterding/RNAseq/blob/master/bin/gtf2bed
