Terminitor

Terminitor is a deep neural network that predicts whether a sequence contains a polyadenylated (poly(A)) cleavage site (CS) at certain position.

For more information, please refer to the preprint: https://www.biorxiv.org/content/10.1101/710699v2

Datasets for download

www.bcgsc.ca/downloads/supplementary/Terminitor

This ftp site contains two datasets, human and mouse, and two corresponding pre-trained models for test.

Dependencies

• Python3
• Numpy
• Keras
• Scikit-learn
• Pybedtools
• Pysam
• HTSeq

A Python environment for these packages can be created with conda, e.g.

conda create --name terminitor pysam pybedtools numpy keras scikit-learn htseq

For more information, consult the user guide for conda.

Train

Usage: train.py [-h] [-v] -polya POLYA -cs CS -non NON -model MODEL -l L

optional arguments:
  -h, --help     show this help message and exit
  -v, --version  show program's version number and exit
  -polya POLYA   Poly(A) CS, fasta file
  -cs CS         Non-poly(A) CS, fasta file
  -non NON       Non-CS, fasta file
  -model MODEL   File name of trained model
  -l L           Length of input sequences

Extract candidate sequence

Usage: extract_from_sequences.py [-h] [-v] -t ANNOT_TRANS -a ANNOT_ALL -m ALN -g GENOME -o O [-u UP_LEN] [-d DOWN_LEN]

optional arguments:
  -h, --help            show this help message and exit
  -v, --version         show program's version number and exit
  -t ANNOT_TRANS, --annot_trans ANNOT_TRANS
                        Transcript annotation file, GTF format. This file
                        contains only transcript level annotation, can be
                        downloaded from the ftp site provided on our Github
                        page
  -a ANNOT_ALL, --annot_all ANNOT_ALL
                        Ensembl annotation file, GTF format. Can be downloaded
                        from Ensembl ftp site
  -m ALN, --aln ALN     The alignment file from assembled transcript contigs
                        to reference genome in BAM format.
  -g GENOME, --genome GENOME
                        Indexed reference genome assembly in FASTA format, which
                        can be downloaded from Ensembl
  -o O                  Output file, fasta format containing candidate
                        sequences to be tested
  -u UP_LEN, --up_len UP_LEN
                        Upstream sequence length
  -d DOWN_LEN, --down_len DOWN_LEN
                        Downstream sequence length

Test

Usage: test.py [-h] [-v] -t TEST_FILE -m MODEL -l L -o OUTPUT

optional arguments:
  -h, --help            show this help message and exit
  -v, --version         show program's version number and exit
  -t TEST_FILE, --test_file TEST_FILE
                        Fasta file to be tested
  -m MODEL, --model MODEL
                        Pre-trained model file
  -l L                  Length of input sequences
  -o OUTPUT, --output OUTPUT
                        Output probabilities

Pipeline

1. For Illumina RNA-seq short reads, run assembly with RNA-Bloom

   java -jar RNA-Bloom.jar -left read2.fq -right read1.fq -revcomp-right -outdir assembly -a 4 -e 1 -stratum 01 -ss -ntcard -fpr 0.005

   For PacBio CCS reads, skip this step

2. Genome alignment with minimap2

   minimap2 -ax splice hg38.mmi rnabloom.transcripts.fa | samtools view -u - | samtools sort -T tmp_prefix -O BAM -o aln.bam
   samtools index aln.bam

3. Extract candidate sequence

   python extract_from_sequences.py -t Homo_sapiens.GRCh38.99.transcripts.gtf -a Homo_sapiens.GRCh38.99.gtf -g Homo_sapiens.GRCh38.dna.primary_assembly.fa -m aln.bam -o extracted_sequences.fa

The GTF file for -a option can be downloaded from Ensembl.

The GTF file for -t option can be generated based on the Ensembl annotation, e.g.

awk '$3=="transcript" {print}' Homo_sapiens.GRCh38.99.gtf > Homo_sapiens.GRCh38.99.transcripts.gtf

The reference genome for -g option can be downloaded from Ensembl and must be indexed, e.g.

samtools faidx Homo_sapiens.GRCh38.dna.primary_assembly.fa

4. Test

   python test.py -t extracted_sequences.fa -m pre_trained_model -l 200 -o probablities.txt