Guidance on Training Augustus for Entamoeba gingivalis

[shubavarshini]

Dear Augustus Team,

I have previously used Augustus to annotate a clinical isolate of Candida auris, which was closely related to Candida albicans. Now, I am working on Entamoeba gingivalis and I have a draft contigs genome assembly put together.

In the species list, I see the non-pathogenic protozoan Leishmania tarentolae, but I am unsure how divergent these two species are. Therefore, I cannot estimate the decrease in sensitivity for gene prediction.

Entamoeba gingivalis is closely related to Entamoeba histolytica, Entamoeba invadens, Entamoeba dispar, and Entamoeba moshkovskii. The whole genomes of these Entamoeba species have been sequenced, and gene annotations are available in NCBI and AmoebaDB.org.

I would like to use these related species to train Augustus and then predict genes for E. gingivalis. However, I am finding the RUNNING_AUGUSTUS.md, README.md and many manuals provided somewhat confusing (and I apologize for that). Any explanation or guidance on the necessary command-line steps for training and then predicting would be extremely helpful.

Thank you for your time and assistance.

[Rong-ao]

Hi @shubavarshini I'm also trying to use AUGUSTUS to annotate two species of genomes without annotation recently and I also felt confused with too complicated informations provided in different docs, including RUNNING-AUGUSTUS.md, retraining.html and README.autoAug.
In my understanding of those information, I think the basic logic of "training" is that, since we have target species without annotation, we can use "closely related species genome annotation" as input to adjust parameters in AUGUSTUS model, then use these parameters, which is efficient for our target species, to predict genes in our target species annotation. Am I right?
(I'm not sure my understanding is 100% correct! Please point out if any mistakes)
Although I still have not figure out how to "train" for my species, I wish you may communicate with me about its usage :)

[shubavarshini]

Hi @Rong-ao. I did not get a notification of your reply. The following are the steps I ended up using to annotate:
#Using augustus v3.5.0 in a conda env
#Folder with model org genomes and training augustus

out_folder=/<define a path>

#Creating a New Species in AUGUSTUS

new_species.pl --species=E_gingivalis_param2

#Merge the files into one training dataset:

cat ${out_folder}/ref1/ref1_training.gb ${out_folder}/ref2/ref2_training.gb ${out_folder}/ref3/ref3.gb ${out_folder}/ref4/ref4_training.gb > ${out_folder}/entamoeba_training.gb

#Too many genes in the training set can overload optimization, especially with a high --kfold value.
#Instead of running optimization on all genes, randomly select a smaller subset:

randomSplit.pl ${out_folder}/entamoeba_training.gb 300
optimize_augustus.pl --species=E_gingivalis_param2 --cpus=8 --kfold=8 ${out_folder}/entamoeba_training.gb.test --onlytrain=${out_folder}/entamoeba_training.gb.train

#Run Initial Training

etraining --species=E_gingivalis_param2 ${out_folder}/entamoeba_training.gb > ${out_folder}/etraining_param2.log 2>&1

#Predict Genes in E. gingivalis (Ab Initio Mode)
#Run AUGUSTUS on the draft genome:

augustus --species=E_gingivalis_param2 contigs_for_gene_prediction.fa --gff3=true --outfile=<path and/or prefix>.gff --errfile=<prefix>.err

#Post-processing
#Getting gene and protein sequences

getAnnoFasta.pl <path and/or prefix>.gff --protein=on --codingseq=on

Hope I was not very late in replying.

[KatharinaHoff]

If you want to train AUGUSTUS on the basis of protein sequences from a closely related species, please consider using Galba. If you have RNA-Seq, please consider using BRAKER.

…

On Fri, Mar 7, 2025 at 10:28 PM Shuba Alampalli ***@***.***> wrote: Hi @Rong-ao <https://github.com/Rong-ao>. I did not get a notification of your reply. The following are the steps I ended up using to annotate: #Using augustus v3.5.0 in a conda env #Folder with model org genomes and training augustus out_folder=/<define a path> #Creating a New Species in AUGUSTUS new_species.pl --species=E_gingivalis_param2 #Merge the files into one training dataset: cat ${out_folder}/ref1/ref1_training.gb ${out_folder}/ref2/ref2_training.gb ${out_folder}/ref3/ref3.gb ${out_folder}/ref4/ref4_training.gb > ${out_folder}/entamoeba_training.gb #Too many genes in the training set can overload optimization, especially with a high --kfold value. #Instead of running optimization on all genes, randomly select a smaller subset: randomSplit.pl ${out_folder}/entamoeba_training.gb 300optimize_augustus.pl --species=E_gingivalis_param2 --cpus=8 --kfold=8 ${out_folder}/entamoeba_training.gb.test --onlytrain=${out_folder}/entamoeba_training.gb.train #Run Initial Training etraining --species=E_gingivalis_param2 ${out_folder}/entamoeba_training.gb > ${out_folder}/etraining_param2.log 2>&1 #Then optimize parameters: This step tunes the gene prediction model to better fit E. gingivalis. optimize_augustus.pl --species=E_gingivalis --cpus=30 --kfold=30 ${out_folder}/entamoeba_training.gb #--aug_exec_dir=${out_folder}/tmp_opt_E_gingivalis #Step 5: Predict Genes in E. gingivalis (Ab Initio Mode) #Run AUGUSTUS on the draft genome: augustus --species=E_gingivalis_param2 contigs_for_gene_prediction.fa --gff3=true --outfile=<path and/or prefix>.gff --errfile=<prefix>.err #Step 6: Post-processing #Getting gene and protein sequences getAnnoFasta.pl <path and/or prefix>.gff --protein=on --codingseq=on Hope I was not very late in replying. — Reply to this email directly, view it on GitHub <#430 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJMC6JB4UUCK3UYMGKOSDYD2TIFOBAVCNFSM6AAAAABW25JIZCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMBXGQ4TGNJXGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***> [image: shubavarshini]*shubavarshini* left a comment (Gaius-Augustus/Augustus#430) <#430 (comment)> Hi @Rong-ao <https://github.com/Rong-ao>. I did not get a notification of your reply. The following are the steps I ended up using to annotate: #Using augustus v3.5.0 in a conda env #Folder with model org genomes and training augustus out_folder=/<define a path> #Creating a New Species in AUGUSTUS new_species.pl --species=E_gingivalis_param2 #Merge the files into one training dataset: cat ${out_folder}/ref1/ref1_training.gb ${out_folder}/ref2/ref2_training.gb ${out_folder}/ref3/ref3.gb ${out_folder}/ref4/ref4_training.gb > ${out_folder}/entamoeba_training.gb #Too many genes in the training set can overload optimization, especially with a high --kfold value. #Instead of running optimization on all genes, randomly select a smaller subset: randomSplit.pl ${out_folder}/entamoeba_training.gb 300optimize_augustus.pl --species=E_gingivalis_param2 --cpus=8 --kfold=8 ${out_folder}/entamoeba_training.gb.test --onlytrain=${out_folder}/entamoeba_training.gb.train #Run Initial Training etraining --species=E_gingivalis_param2 ${out_folder}/entamoeba_training.gb > ${out_folder}/etraining_param2.log 2>&1 #Then optimize parameters: This step tunes the gene prediction model to better fit E. gingivalis. optimize_augustus.pl --species=E_gingivalis --cpus=30 --kfold=30 ${out_folder}/entamoeba_training.gb #--aug_exec_dir=${out_folder}/tmp_opt_E_gingivalis #Step 5: Predict Genes in E. gingivalis (Ab Initio Mode) #Run AUGUSTUS on the draft genome: augustus --species=E_gingivalis_param2 contigs_for_gene_prediction.fa --gff3=true --outfile=<path and/or prefix>.gff --errfile=<prefix>.err #Step 6: Post-processing #Getting gene and protein sequences getAnnoFasta.pl <path and/or prefix>.gff --protein=on --codingseq=on Hope I was not very late in replying. — Reply to this email directly, view it on GitHub <#430 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJMC6JB4UUCK3UYMGKOSDYD2TIFOBAVCNFSM6AAAAABW25JIZCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMBXGQ4TGNJXGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Rong-ao]

@shubavarshini Thanks a lot for your reply! Never too late! I'm trying to use the pipeline you provided (It really helps me much, thanks again) and also Galba or BRAKER as @KatharinaHoff mentioned. I wonder whether you got your ideal annotation results, and did you also try these two softwares?

[shubavarshini]

@Rong-ao

I only have DNA long read sequencing to assemble the genome. Since I have no supporting data like RNA-seq, I can't use Galba or BRAKER. Rather I do not think I can optimize them for a new organism. I might be wrong not sure.

I'm still analyzing downstream if my annotation results turned out fine with homology searches. So cannot for sure say if Augustus did give me "ideal annotation results".

[KatharinaHoff]

How is the repeat content of your genome, and what is the genome size? What is the closest annotated genome?

…

On Wed, Mar 12, 2025 at 2:49 PM Shuba Alampalli ***@***.***> wrote: I only have DNA long read sequencing to assemble the genome. Since I have no supporting data like RNA-seq, I can't use Galba or BRAKER. Rather I do not think I can optimize them for a new organism. I might be wrong not sure. I'm still analyzing downstream if my annotation results turned out fine with homology searches. So cannot for sure say if Augustus did give me "ideal annotation results". — Reply to this email directly, view it on GitHub <#430 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJMC6JC7VM3AIA23RWK6IUL2UA3OVAVCNFSM6AAAAABW25JIZCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMJXHE2DSNBRGA> . You are receiving this because you were mentioned.Message ID: ***@***.***> [image: shubavarshini]*shubavarshini* left a comment (Gaius-Augustus/Augustus#430) <#430 (comment)> I only have DNA long read sequencing to assemble the genome. Since I have no supporting data like RNA-seq, I can't use Galba or BRAKER. Rather I do not think I can optimize them for a new organism. I might be wrong not sure. I'm still analyzing downstream if my annotation results turned out fine with homology searches. So cannot for sure say if Augustus did give me "ideal annotation results". — Reply to this email directly, view it on GitHub <#430 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJMC6JC7VM3AIA23RWK6IUL2UA3OVAVCNFSM6AAAAABW25JIZCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMJXHE2DSNBRGA> . You are receiving this because you were mentioned.Message ID: ***@***.***>