Heuristic tool for pairing up reverse complement reads from fasta/fastq files.
Sniff loads sequences in batches displaying progress along the way. Once all sequences have been loaded into memory they are again processed in batches. Each batch is used for constructing a target index from reverse complemented reads. Original reads are then mapped against the constructed index using a seed and chain approach with minor modifications. For each read we remember the strongest matching reverse complement read and output reverse complement pairs as overlaps. Here we define an overlap as a tuple query_name, query_start, query_end, target_name, target_start, target_end. Later those overlaps are processed with a pre-trained machine learning model outputting the final result in a csv/tsv format for later use.
| precision | support |
|---|---|
| 0.809323 | 42024 |
Here we consider a pair to be a true positive if reads map with different strands to an approximately the same area on the reference. We say that the approximate mapping is sufficient if the intersection of covered area by both reads divided by the union of the same is larger than 0.9.
git clone git@github.com:tbrekalo/sniff.git
cd sniff
make releaseFrom sniff root directory:
source ./venv/bin/activate
./build/bin/sniff -t 32 path_to_reads.fasta > /tmp/sniff.csv
python ./scripts/inference/lgbm_filter.py -m resources/sniff-lgbm-model.pkl -o /tmp/sniff.csv > pairs.csv- linux kernel 2.6.32 or higher
- gcc 11 or higher
- clang 11 or higher
- intel tbb 2020.3
- should be compatible with later oneapi versions
- conan2 with configured profile
- cmake 3.21 or higher
- git 2.25.1 or higher
- earlier version should do just fine
- git is required for cmake to fetch par of internal dependencies
- Catch2
- fetched via cmake if missing
- conan2==0.0.4
- joblib==1.3.2
- lightgbm==4.1.0
- polars==0.18.5
- psutil==5.9.5
- pydantic==1.10.9
- scikit-learn==1.3.1