A tool for metagenomics assemblies polishing, especially for Nanopore long read assemblies

1. download conda from https://docs.anaconda.com/miniconda/ ; run conda init and reopen shell terminal
2. clone this project
3. At project directory, run bash install.bash

The conda environment metaconda will be created and the metaconnet is in the conda env.

# if an error like "conda init before activate" appears
# just run bash line by line
conda env create -n metaconda -f env.yml # to create conda environment
conda activate metaconda # activate the newly installed environment
pip install . # install the current version of metaconnet by source code 

If samtools has been unsuccessful to be installed, check the conda env channel order.
The channel order to correctly install is

channels:
  - conda-forge
  - bioconda
  - defaults

• minimap2
• samtools
• python3.8 or python 3.9
  Make sure samtools, minimap2 and python are at the bin path

1. Clone this project
2. At project directory, run pip install -r requirements.txt
3. At project directory, run pip install .

usage: MetaCONNET [-h] [--sr SR SR] --lr LR --c C --o O --n N [--t T] [--v] [--fc FC]

MetaCONNET input

optional arguments:
  -h, --help            show this help message and exit
  --sr SR SR            NGS read fastq/fasta read files
  --lr LR               long read fastq/fasta file
  --c C                 contig fastq/fastq file
  --o O                 output folder
  --n N                 task name
  --t T                 thread number
  --v, --version        show program's version number and exit
  --fc, --flowcell      flow cell version : R10, R9. if not given, R9 flow cell model will be used

--lr and --c are required. If you want to combine short reads, please use --sr with read1 and read2 file path.
You can use --o to specify a working directory.
You will need to specify a name for the task as the future prefix for the output file by --n.
MetaCONNT has allowed multithreading, so please remember to include thread numbers --t for parallel. Version 1.1 has added R10 model, specify --fc as R10 to use R10 models

These directories and files will be created during polishing at the working directory.

• data
• round1
• round2
• <task_name>_polished.fasta

The polished fasta is in <task_name>_polished.fasta as a softlink to last fasta in round2 recovery phase

For testing purposes, you can cd to the test folder and execute bash test.sh. The test data read : longreads.fastq and contigs: contigs.fasta are also in the test folder.

For who is interested in training a model of their own, first of all, the models are under ~/pipeline/training_model .

The training codes are under ~/train

Step 1 : Prepare training data.

Run prepare_training.sh

Example :

#corretion
wdir=phase1
bam=map.bam # reads to assembly bam
assembly=assembly.fasta
ref=ref.fasta
time bash prepare_training.sh $wdir $ref  $bam $assembly 0 dataset_name
#recovery
wdir=phase2
time bash prepare_training.sh $wdir $ref  $bam $assembly 1 dataset_name

Step 2 : Run training.
This requires you to have a gpu and a tensorflow-gpu ready environment

Example :

model1=meta_correction
model2=meta_recovery
#corretion
metaconnet_train -epochs 200 -parallel phase1/dataset_name.parallel -train phase1 -o $model1 -phase correction
#recovery
metaconnet_train -epochs 200 -parallel phase2/dataset_name.parallel -train phase2 -o $model2 -phase recovery

This will train 200 epochs of the training datasets and save the final epoch model and each epoch checkpoint. You can load the final model and find the best model checkpoint and save to keras models.