Parrelizable codeml?

dnds_one.sh

#for seq in alignments/ENSG*.best.fas
#do 
    # There is some evidence that forcing trees to fit the spp tree will inflate
    # dN/dS estimates. Better to estimate the tree every time. At the same time,
    # label the human branch and delet branch lengths which aren't useful
gene=`basename  ${seq%%.*}`
raxml -s ${seq} -T 2 -p 1231 -m GTRGAMMA -n ${gene}_delete_me
sed -e "s/:0.[0-9]*//g"  -e "s/sapiens/sapiens#1/g" RAxML_bestTree.${gene}_delete_me >${gene}.tr
rm *.${gene}_delete_me
    
    #now run codeml using the template control file update with the current seq
echo -e 'seqfile = '${seq}'\n treefile = '${gene}'.tr\n outfile = '${gene}'_branch.txt' | cat - codeml_branch_template.ctl  > ${gene}_branch.ctl
echo -e 'seqfile = '${seq}'\n treefile = '${gene}'.tr\n outfile = '${gene}'_null.txt'   | cat - codeml_null_template.ctl    > ${gene}_null.ctl
codeml ${gene}_branch.ctl
codeml ${gene}_null.ctl

    #get the good stuff from the results files
scripts/extract_codeml.py ${gene} >> results/codeml_res.csv
    #including the number of non-synonymous changes in the human branch
#grep "(homo_sapiens)" rst | sed  "s/.*n=\([ 0-9]*.[0-9]*\).*/${gene}\t\1/"  >> results/num_non_syn.tsv

    #and clean up the god-awful mess it makes
#rm 2NG.*
#rm 4fold.nuc
#rm lnf
#rm rst*
#rm rub
# ...any my mess too
rm ${gene}.tr
rm ${gene}*.ctl
rm ${gene}_branch.txt
rm ${gene}_null.txt