NanoTS

NanoTS is a deep-learning-based variant caller for SNP detection from Nanopore transcriptome sequencing data. It provides efficient and accurate SNP calling from long-read Nanopore RNA sequencing data, supporting both cDNA and direct RNA technologies, particularly with the latest R10.4 chemistry.

---

Installation

Note: NanoTS is currently supported only on Linux systems.

Option 1: Conda

Ensure you have Conda installed, then create and activate the nanoTS environment:

git clone [email protected]:Xinglab/NanoTS.git # clone nanoTS
cd nanoTS
conda env create -f environment.yml # create nanoTS conda environment
conda activate nanoTS

Install nanoTS:

pip install .

Alternatively, run the NanoTS runner directly:

python ./source/nanoTS-runner.py [OPTIONS]

Option 2: Singularity

Ensure you have Singularity installed, then pull the nanoTS container:

singularity pull library://zelinliu/nanots/nanots:latest

---

Usage

NanoTS is a command-line tool with the following subcommands:

Subcommand	Description
bam	Suffix each BAM alignment QNAME by count per read.
unphased_call	Extracts candidate SNPs and features from a BAM file.
haplotype	Performs haplotype phasing using a VCF, reference, and BAM file.
phased_call	Processes phased SNP calls with the deep-learning model.
clean	Removes temporary files from the output directory.
full_pipeline	All-in-one command that includes all of the above steps.

General Command Format

nanoTS <subcommand> [OPTIONS]

---

Models

NanoTS uses pre-trained models for SNP calling:

Model Name	Description
unphased_cDNA_nanopore_R104_HG002.pth	Unphased SNP model for cDNA
unphased_dRNA_nanopore_R104_HG002.pth	Unphased SNP model for direct RNA
phased_cDNA_nanopore_R104_HG002.pth	Phased SNP model for cDNA
phased_dRNA_nanopore_R104_HG002.pth	Phased SNP model for direct RNA

---

Subcommands & Arguments

1️⃣ bam

Append a numeric suffix to each BAM alignment QNAME based on its count per read, ensuring each alignment has a unique QNAME (e.g., read_1, read_2).

Required Arguments:

Argument	Description
-i/--input	Input sorted BAM file.
-o/--output	Output BAM file with suffixed QNAMEs.

Optional Arguments:

Argument	Default	Description
--region	None	Target region (chr1 or chr1:1000-2000).

2️⃣ unphased_call

Extracts candidate variants and features for deep learning-based SNP calling.

Required Arguments:

Argument	Description
--bam	Sorted and indexed BAM file.
--ref	Reference genome FASTA file.
--outdir	Output directory.
--model	Pre-trained model (.pth file).

Optional Arguments:

Argument	Default	Description
--threads	24	Number of CPU threads.
--region	None	Target region (chr1 or chr1:1000-2000).
--ALT	2	Minimum ALT SNP coverage.
--total	2	Minimum total base coverage.
--ratio	0.05	Minimum ALT allele frequency.
--depth	1000	Maximum reads per variant (0 to disable limit).

🔹 Note: Increasing --ALT and --total thresholds may affect haplotype inference efficiency.

---

3️⃣ haplotype

Performs haplotype phasing using an input VCF, reference genome, and BAM file.

Required Arguments:

Argument	Description
--vcf	Input VCF file.
--ref	Reference genome FASTA file.
--bam	Input BAM file.
--outdir	Output directory.

Optional Arguments:

Argument	Default	Description
--hap_qual	10	QUAL threshold to filter SNPs during haplotype phasing.

---

4️⃣ phased_call

Processes phased variant calls using the deep-learning model.

Required Arguments:

Argument	Description
--bam	Sorted and indexed BAM file.
--ref	Reference genome FASTA file.
--outdir	Output directory.
--model	Pre-trained model (.pth file).

Optional Arguments:

Argument	Default	Description
--threads	24	Number of CPU threads.
--depth	1000	Maximum reads per variant (0 to disable limit).

---

5️⃣ clean

Removes temporary files in the output folder.

Required Arguments:

Argument	Description
--outdir	Output directory to clean.

---

6️⃣ full_pipeline

Runs the entire NanoTS pipeline from a BAM file through final phased VCF output, then cleans up intermediate files.

Required Arguments:

Argument	Description
--bam	Input sorted & indexed BAM file.
--ref	Reference genome FASTA file.
--model_unphased	Path to unphased model (.pth file).
--model_phased	Path to phased model (.pth file).
--outdir	Output directory for all steps.

Optional Arguments:

Argument	Default	Description
--region	None	Target region (e.g., chr1 or chr1:1000-2000).
--threads	24	Number of CPU threads.
--ALT	2	Minimum ALT SNP coverage.
--total	2	Minimum total base coverage.
--ratio	0.05	Minimum ALT allele frequency.
--depth	1000	Maximum reads per variant (0 to disable limit).
--hap_qual	10	QUAL threshold to filter SNPs during haplotype phasing.

---

Example Workflow

Here’s an end-to-end example using NanoTS:

cd example/

# Download and prepare the reference genome
wget https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/hg38.fa.gz
gzip -d hg38.fa.gz
samtools faidx hg38.fa

# Align Nanopore reads to the reference genome
minimap2 -ax splice -ub -t 24 -k 14 -w 4 --secondary=no hg38.fa tutorial.fastq.gz | \
  samtools sort -o tutorial.bam
samtools index tutorial.bam
# Define output directory
outdir=nanoTS_result/
##################################
#### All-in-one step ####
nanoTS full_pipeline \
  --bam tutorial.bam \
  --ref hg38.fa \
  --model_unphased ../model/unphased_cDNA_nanopore_R104_HG002.pth \
  --model_phased   ../model/phased_cDNA_nanopore_R104_HG002.pth \
  --outdir $outdir \
  --threads 24 \
  --ALT 2 \
  --total 2 \
  --ratio 0.05 \
  --depth 1000 \
  --hap_qual 10
##################################
#### Separate steps ####
# Step 1: Suffix each BAM alignment QNAME by count per read
nanoTS bam \
  -i tutorial.bam \
  -o tutorial.qname.bam

# Step 2: Unphased SNP calling
nanoTS unphased_call \
  --bam tutorial.qname.bam \
  --ref hg38.fa \
  --threads 24 \
  --ALT 2 \
  --total 2 \
  --ratio 0.05 \
  --depth 1000 \
  --outdir $outdir \
  --model ../model/unphased_cDNA_nanopore_R104_HG002.pth

# Step 3: Haplotype phasing
nanoTS haplotype \
  --vcf ${outdir}unphased_predict.pass.vcf \
  --ref hg38.fa \
  --bam tutorial.qname.bam \
  --outdir $outdir \
  --QUAL 10

# Step 4: Phased SNP calling
nanoTS phased_call \
  --bam tutorial.qname.bam \
  --ref hg38.fa \
  --threads 24 \
  --depth 1000 \
  --outdir $outdir \
  --model ../model/phased_cDNA_nanopore_R104_HG002.pth

# Step 5: Clean temporary files
nanoTS clean \
  --outdir $outdir

Run NanoTS with Singularity (always bind host folders you read/write)

cd example/

SIF=../../../nanots_latest.sif # use your SIF path
EXAMPLE_DIR="$(pwd)"
MODEL_DIR="$(cd ../model && pwd)"   # the model folder in this repository. Adjust if your model files live elsewhere

OUTDIR="$EXAMPLE_DIR/nanoTS_result"
mkdir -p "$OUTDIR"

# Bind BOTH the data dir and the model dir so the container can see them.
singularity exec -B "$EXAMPLE_DIR","$MODEL_DIR" "$SIF" \
  /opt/conda/envs/nanoTS/bin/nanoTS full_pipeline \
    --bam "$EXAMPLE_DIR/tutorial.bam" \
    --ref "$EXAMPLE_DIR/hg38.fa" \
    --model_unphased "$MODEL_DIR/unphased_cDNA_nanopore_R104_HG002.pth" \
    --model_phased  "$MODEL_DIR/phased_cDNA_nanopore_R104_HG002.pth" \
    --outdir "$OUTDIR" \
    --threads 24 \
    --ALT 2 --total 2 --ratio 0.05 --depth 1000 --hap_qual 10

---

Output Files

• unphased_predict.pass.vcf → SNP calls generated from the unphased_call step.
• phased_predict.pass.vcf → SNP calls generated from the phased_call step after haplotype phasing.

All results are stored in the specified --outdir directory.

---

License

This project is licensed under the GPL-3.0 License. See the LICENSE file for details.

---

Contact

For questions, bug reports, or feature requests, contact:

📧 Zelin Liu – [email protected]