MASON - Help

FAQ's

Find problems which are often encountered when using MASON

How to find and download a genome/annotation file in NCBI?

To upload a custom organism to MASON, the genome sequence and annotation files must be provided in FASTA and GFF format, respectively. Below we describe an example explaining how to find and download these files for the pathogenic bacterium Campylobacter jejuni:

Open the NCBI website and select “Nucleotide” in the dropdown menu
Add the name of your target organism (here, “Campylobacter jejuni subsp. jejuni NCTC 11168”) and click the search button
Once the query results are shown, select the complete genome of your target organism and click on it
Click “Send to” to download available files for the organism
Select the “Complete Record button”
Click on the “File” button
Select “FASTA” from the dropdown menu
Click “Create File” to download the FASTA file of the complete genome
Repeat steps 4-8, but select “GFF3” instead of “FASTA” at step 7 to download the annotation file in GFF format

How to find the locus tag of my selected target gene in the target organism?

MASON requires locus tags as identifiers for target genes, because they are more consistent across species than gene names. Below we describe an example explaining how to find the locus tag of the essential gene “pheA” in Campylobacter jejuni:

Open the GFF annotation file in your favourite text editor
Use the search and find utility of your text editor. Often, CTRL + F opens up the “Find” tool. Now enter the name of your target gene (here, “pheA”) and press ENTER
In the 9th column of the line, information of the gene is stored. Find the “locus_tag=” field and copy the locus tag
Add the copied locus tag to MASON as a target gene

How to fill in the start form?

Below each start form is a short description of what to fill in. We show an example on a case in which a user uses MASON to find ASO sequences for one essential gene (accC) and another, non-essential gene (b3256) in E. coli K12:

The custom name can be any name, here we use “test”
We use the pre-selectable genome of E. coli K12. Likewise, other genomes can be selected or a custom genome can be uploaded, see the other help page for information on this
We select the essential gene “accC”
We additionally select another gene, with the locus tag “b3258”. This is optional and can be left out if already a gene in the dropdown menu was selected
The length of the ASOs is chosen to be 10 to get 10mer ASOs
2 mismatches are allowed to be inside the ASOs to be considered a mismatch
We want to design only sequences targeting the start codon. Therefore, this option is left empty
No other genome is selected for off-target screening. Optionally, we could have added the human genome or the microbiome to be screened for off-targets
The start button can now be clicked to start the MASON process

How does the machine-learning option work?

If the "Use machine-learning (random forest) model to predict MICs" box is ticked, MASON feeds each designed ASO into a random forest regressor and returns a predicted minimum inhibitory concentration (MIC) along with a relative rank.

The model was trained on an internal, unpublished dataset of ~586 PNAs targeting the essential genome of uropathogenic E. coli str. 536 (UPEC 536). Each PNA in the training set was assayed for its MIC against UPEC 536.
Input features include PNA sequence properties (Tm, % self-complementary bases, purine percentage), the number of off-targets in TIR regions of UPEC 536 (1 mismatch), the codon-adaptation index (CAI) of the target gene, and the minimum free energy (MFE) of the target's translation initiation region.
Predicted MIC values are reported in µM. Because the model was trained on UPEC 536 data, they should not be interpreted as absolute MICs in other organisms — instead, use the MIC_rank column to compare ASOs relative to each other.

What is a SHAP value?

SHAP (SHapley Additive exPlanations) values explain individual predictions by attributing each feature a contribution that pushes the model output above or below the average prediction (the "base value"). They are based on game-theoretic Shapley values and are computed here with the shap Python library applied to the random forest MIC predictor.

For every designed ASO we generate a force plot which can be opened by hovering over the SHAP link in the MIC_predicted column of the result table. In the plot:

the bold value on the axis is the model's predicted MIC for that ASO,
the "base value" is the average predicted MIC across the training set,
orange bars push the prediction up (worse predicted MIC), dark blue bars push it down (better predicted MIC), and the size of the bar reflects how strongly that feature moved the prediction.

This makes it possible to see, for any given ASO, which sequence properties or off-target counts are responsible for the predicted MIC.

How to interpret the off-target output table?

Below each output column is a short description of what it means. Each row denotes an off-target match with a specific gene of a respective PNA.

locus_tag: Denotes locus tag of off-target gene
gene_name: Denotes gene name of off-target gene
strand: Denotes strand of off-target match (target mRNA)
trans_coord: Denotes the location of off-target match. I.e. first matching base respective to start codon of off-target gene
off_target_seq_mRNA: Denotes sequence of mRNA of organism that is subject to an off-target effect
probe_id: Unique identifier of off-target. Consists of locus tag (target gene), gene name (target gene) and ASO name
mRNA_target_seq: Denotes targeted sequence of ASO. I.e. the target mRNA if there are no mismatches
num_mismatch: Number of mismatches of off-target
mismatch_positions: Positions of mismatches
longest_stretch: length of longest matching stretch without a mismatch
matching_sequence: longest matching stretch without a mismatch (ASO-mRNA, mRNA sequence shown)
ASO: Unique name of the ASO
TIR: Whether the off-target is within the translation initiation region of a gene

Contact