FlaHMM: unistrand flamenco-like piRNA cluster prediction in Drosophila species
FlaHMM can be cloned from our GitHub repository using the following command
git clone https://github.com/Hannon-lab/FlaHMM.git
To install all Python dependencies, we recommend creating a conda environment and install them using pip (or pip3, if applicable)
conda create -n FlaHMM python=3.8
conda activate FlaHMM
pip install pandas
pip install scikit-learn
pip install "hmmlearn==0.2.7"
pip install matplotlib
pip install tqdm
pip install seaborn
pip install openpyxl
FlaHMM has primarily been tested using hmmlearn v0.2.7. If needed, FlaHMM can also run with hmmlearn 0.2.8 or newer, however, please note that MultinomialHMM has been renamed to CategoricalHMM.
FlaHMM requires information about Gypsy transposon content across the target genome. The easiest option will be to download our pre-computed transposon content files using the following command.
cd FlaHMM
wget https://content.cruk.cam.ac.uk/ghlab/Susanne/FlaHMM/bins.tar.gz
tar xf bins.tar.gz
This includes pre-processed files for 193 genome assemblies from 119 species. All full list of all pre-processed genome assemblies and species is available here. Alternatively, you can prepare custom transposon content annotations using the instructions below.
To run FlaHMM, go to the FlaHMM directory and run FlaHMM.py with the following parameters:
- Drosophila species genome assembly to make predictions on (e.g., Dmel.dm6, Dfic.GCF_018152265 or Dper.d101g; please find a full list here).
- Bin size in kb (2.5, 5 or 10)
- LTR/Gypsy threshold (0.025, 0.05, 0.075, 0.1 or 0.2)
For example:
python FlaHMM.py --species Dmel.dm6 --bins 5 --threshold 0.025
Results are saved in results folder and can be visualised with the Results_Visualisation.ipynb Jupyter notebook found in the supplementary repository.
Running FlaHMM with the command above would give the following results
$ python FlaHMM.py --species Dmel.dm6 --bins 5 --threshold .025
=========================================================================================
FlaHMM v1.0.0
A tool for flam-like cluster predictions based on LTR/Gypsy transposon content.
If you encounter any issues or have any suggestions, please visit our GitHub issues page:
https://github.com/Hannon-lab/FlaHMM/issues
=========================================================================================
Formatting test data....
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 29918/29918 [00:02<00:00, 11260.50it/s]
Making flaHMM predictions for Dmel.dm6...
Done. Predictions saved in results folder as results/results_Dmel.dm6_Bin_5k_threshold_0.025.txt.
Here, the output file results/results_Dmel.dm6_Bin_5k_threshold_0.025.txt is indicated, and has the following structure
$ head results/results_Dmel.dm6_Bin_5k_threshold_0.025.txt
chr bin_start bin_end strand species_test region_binary emission_0.025 pred_0.025 proba_NoCluster_0.025 proba_Cluster_0.025 proba_Centromere_0.025
0 chr2L 0 5000 plus Dmel.dm6 0 0 0 0.9999719472651957 7.436615274089606e-06 2.0616119612784525e-05
1 chr2L 5000 10000 plus Dmel.dm6 0 0 0 0.9999847199204476 2.347948548920461e-06 1.2932131042633266e-05
2 chr2L 10000 15000 plus Dmel.dm6 0 0 0 0.9999911049428629 7.586888007424614e-07 8.13636828857453e-06
3 chr2L 15000 20000 plus Dmel.dm6 0 0 0 0.9999945944408082 2.6234102246346134e-07 5.143218177404513e-06
4 chr2L 20000 25000 plus Dmel.dm6 0 0 0 0.9999966175533153 1.0732452310906451e-07 3.2751220558534618e-06
5 chr2L 25000 30000 plus Dmel.dm6 0 0 0 0.9999978318904452 5.8910512990356805e-08 2.109198985609179e-06
6 chr2L 30000 35000 plus Dmel.dm6 0 0 0 0.9999985746911687 4.378999174827752e-08 1.3815187994768812e-06
7 chr2L 35000 40000 plus Dmel.dm6 0 0 0 0.9999990335761717 3.9067539081585073e-08 9.273563859768871e-07
8 chr2L 40000 45000 plus Dmel.dm6 0 0 0 0.9999993185044835 3.759258416502015e-08 6.439028927488428e-07
Specifically, the output data contains the following columns:
- chr - Chromosome
- bin_start - Start coordinate
- bin_end - End coordinate
- strand - Strand
- species_test - Name of species (and assembly)
- region_binary - Known annotation if available (0 = none, 1 = flam(like), 2 = centromere)
- emission_0.025 - Emission (0 = TEs below threshold, 1 = TEs on one strand, 2 = TEs on both strands)
- pred_0.025 - Predicted region (0 = none, 1 = flam(like), 2 = centromere)
- proba_NoCluster_0.025 - Predicted probability for none (0)
- proba_Cluster_0.025 - Predicted probability for flam(like) (1)
- proba_Centromere_0.025 - Predicted probability for centromere (2)
To run FlaHMM, you need to calculate transposon content for the genome of interest using RepeatMasker with a suitable transposon library (FASTA). If no transposon library is available for your species of interest, we recommend running EDTA. Alternatively, you could use an existing library (e.g., RepBase) from a closely related species. Detailed instructions are found below:
- examples/run_EDTA - Scripts/instructions to generate EDTA transposon libraries (optional)
- examples/make_bins - Scripts to calculate LTR/Gypsy content genome-wide per 2.5 kb, 5 kb, or 10 kb bins from RepeatMasker output
If you use your own BED-like input files, put them into the bins directory using the following directory structure
$ tree bins
bins
├── bins_10k
│ ├── minus
│ │ └── Dmel.dm6.bed
│ ├── plus
│ │ └── Dmel.dm6.bed
└── bins_5k
├── minus
│ └── Dmel.dm6.bed
└── plus
└── Dmel.dm6.bed
and ensure that the BED-like files have the following format.
If you find FlaHMM useful for your research, please consider citing it:
FlaHMM: unistrand flamenco-like piRNA cluster prediction in Drosophila species using hidden Markov models
Maria-Anna Trapotsi, Jasper van Lopik, Gregory J Hannon, Benjamin Czech Nicholson, Susanne Bornelöv
bioRxiv 2024.05.14.592433; doi: https://doi.org/10.1101/2024.05.14.592433