This tutorial uses FABIAN-variant to analyse human variants from a whole-genome sequencing (WGS) run and explains how to interpret the results. FABIAN-variant also supports mouse genomes (mm10 and mm39), which can be selected via the organism icons on the homepage.
The sample results from this tutorial can be viewed and downloaded at this link without repeating the analysis:
https://fabianapp.org/variant26/1783324764_77484
Contents
Select variants
Download one of the sample files:
- VCF_Sample_File_WGS_GRCh37_chr1.zip (~1.6 MB, ~310k variants) – recommended for this tutorial
- VCF_Sample_File_WGS_GRCh37.zip (~20 MB, ~4M variants) – full genome
These files contain a public WGS VCF with a disease-causing variant spiked into the PKLR gene promoter (PMID: 11054094). It is good practice to zip or gzip large VCF files before upload to FABIAN-variant.
Open the FABIAN-variant homepage at https://fabianapp.org/variant26/ and click on 'VCF file' in the menu on the left side.
In the 'VCF file' input field, upload the sample file and select 'GRCh37/hg19'.
Below 'Variant filters', click on 'Variants near selected genes (select to enter)'. A box appears for filtering the VCF to regions near the selected genes. Enter PKLR:
The minimum coverage filter can exclude positions with low read depth, which are usually unreliable. Variants present in gnomAD can also be excluded from the analysis. Both filters are disabled by default. For this tutorial, leave them disabled.
Select transcription factors
FABIAN-variant currently supports 1,530 human transcription factors (TFs) and 1,123 mouse TFs. Click on 'Select individually...' to choose the TFs to include in the analysis. The 'Search by name' field and the 'Select by first letter' buttons assist with selection.
'Known TFBSs' searches for transcription factors shown to bind at the variant positions, based on ENCODE, ReMap 2022, ChIP-Atlas, UniBind, and Ensembl Regulation data. The checkboxes below control which databases are included in this search. ENCODE is selected by default.
Select models
FABIAN-variant implements 40,315 models for human TFs. These include transcription factor flexible models (TFFMs) and position weight matrices (PWMs) pooled from multiple publicly accessible data sources. In addition, 1,259 deep learning models (BPNet) can optionally score variants when genomic coordinates are provided. Unlike PWMs, TFFMs capture dependencies between adjacent positions.
The model sources are organized into two sections: TFFMs (currently JASPAR 2024) and PWMs (from JASPAR 2026, HOCOMOCO v14, and other databases). Below the model sources, the 'Options' section contains two additional settings: 'Show deep learning predictions' adds BPNet predictions when genomic coordinates are provided, and 'Same species models only' excludes models derived from other species.
In this tutorial, use the 'none' link next to 'PWMs' to unselect all PWM sources, and uncheck 'Show deep learning predictions'. This leaves 1,282 TFFM models to be included in the analysis. Click on 'Analyse'.
Computation
While the analysis runs, FABIAN-variant reports progress in the 'computation log'. Filter options are disabled during this time.
When computation finishes, the log disappears. It can be restored by checking the 'Options > Show log' checkbox in the top menu.
Results basics
The results show variants near the PKLR gene and the transcription factors with known binding sites in this region.
Checkboxes and radio buttons in the top menu refine and sort the results table. To keep only transcription factors with a notable predicted effect on binding, check 'Score...' in the 'Filter TFs' options and confirm the suggested minimum of 0.05.
Changes are reflected immediately in the results table:
To focus on a single variant, click on 'Filter variants > Select ...'. The question mark next to it explains how selection filtering works.
Now click on variant chr1:155271258T>C.12 so that all other variants have a grey background (selected state).
Press the Backspace key to hide all selected variants.
Now only the single variant chr1:155271258T>C.12 is displayed in the table.
Filtering and sorting apply only to visible TFs and variants.
In the 'Filter TFs' options, uncheck 'Known TFBSs only'. Then, in the 'Sort TFs' options, click on 'Binding loss'.
When only a few variants are displayed, FABIAN-variant automatically splits the table into multiple columns (above: 6 columns). To display results in a single column, uncheck 'Options > Panel layout'.
Details view
Hovering over any table cell displays the model scores and the positions of the identified motifs relative to the variant.
FABIAN-variant predicts that the variant chr1:155271258T>C results in loss of GATA1 binding. The prediction is based on the available models, which score the reference sequence much higher than the variant sequence. If a deep learning model is available for the TF, its prediction is also shown. See the documentation for scoring and the evaluation of models.
Clicking the table cell opens the details in a separate window (e.g., for printing).
Downloading results
FABIAN-variant provides several download options. This example downloads the complete results file containing all scores. On the results page, click on 'Download: All results'. The format is described in the documentation.
Another disease-related example
For a further example, see an ACKR1 promoter variant that disrupts GATA1 binding:
https://fabianapp.org/variant26/#single-ex
Contact
If you have suggestions about this software, please email robin.steinhaus (at) bih-charite.de. If you discover a bug, please submit a ticket via email using this link.