noise2read 0.3.0

Turn ‘noise’ to signal: accurately rectify millions of erroneous short reads through graph learning on edit distances

noise2read, originated in a computable rule translated from PCR erring mechanism that: a rare read is erroneous if it has a neighboring read of high abundance, turns erroneous reads into their original state without bringing up any non-existing sequences into the short read set(&lt 300bp) including DNA and RNA sequencing (DNA/RNA-seq), small RNA, unique molecular identifiers (UMI) and amplicon sequencing data.

Click noise2read to jump to its documentation

Note: All the experimental results obtained in this study utilised version 0.2.7 of noise2read.

Quick-run example

Quick-run example for testing noise2read by setting only 1 trial for Optuna and 10 estimators for xGboost which are not the parameters used in our paper.

• noise2read installation

Please refer to QuickStart or Installation.

• Clone the codes with datasets in github

git clone https://github.com/Jappy0/noise2read
cd noise2read/Examples/simulated_miRNAs

• Quick-run testing noise2read on D14

  • with high ambiguous errors correction and using GPU for training (running about 4 mins with 26 cores and GPU)

  noise2read -m correction -c ../../config/quick_test.ini -a True -g gpu_hist

Examples for correcting simulated miRNAs data with mimic UMIs by noise2read

Take data sets D14 and D16 as examples.

• noise2read installation

Please refer to QuickStart or Installation.

• Clone the codes with datasets in github

git clone https://github.com/Jappy0/noise2read
cd noise2read/Examples/simulated_miRNAs

• Reproduce the evaluation results for D14 and D16 from raw, true and corrected datasets

noise2read -m evaluation -i ./raw/D14_umi_miRNA_mix.fa -t ./true/D14_umi_miRNA_mix.fa -r ./correct/D14_umi_miRNA_mix.fasta -d ./D14
noise2read -m evaluation -i ./raw/D16_umi_miRNA_subs.fa -t ./true/D16_umi_miRNA_subs.fa -r ./correct/D16_umi_miRNA_subs.fasta -d ./D16

• correcting D14

  • with high ambiguous errors correction and using GPU for training

  noise2read -m correction -c ./configs/D14.ini

  • without high ambiguous errors correction and using GPU for training

  noise2read -m correction -c ./configs/D14_without_high.ini

• correcting D16

  • with high ambiguous errors correction and using GPU for training

  noise2read -m correction -c ./configs/D16.ini

  • without high ambiguous errors correction and using GPU for training

  noise2read -m correction -c ./configs/D16_without_high.ini

• Expected Results

Please find the expected log files and correction results at the folder noise2read of benchmark for correcting data sets of D14-D16. The results under noise2read and noise2read-1 represent the corrected results with and without high ambiguous errors’ prediction, respectively.

Note: Noise2read may produce slightly different corrected result from these results under Examples/simulated_miRNAs/correct and correction. This is because the easy-usable and automatic tuning of the classifiers’ parameters facilitates wide-range explorations, different best models are obtained for each training, but the final prediction results are stable within a certain range. We have discussed this in the Discussion section of our paper.

Examples for correcting outcome sequence of ABEs and CBEs by noise2read

• Clone the codes

git clone https://github.com/Jappy0/noise2read
cd noise2read/CaseStudies
mkdir ABEs_CBEs
cd ABEs_CBEs

• Download datasets under the folder of data of D32_D33.

• Using noise2read to correct the datasets. The running time of each experiment is about 13 minutes using 26 cores and GPU for training.

noise2read -m correction -i ./data/D32_ABE_outcome_seqs.fasta -a False -d ./ABE/
noise2read -m correction -i ./data/D33_CBE_outcome_seqs.fasta -a False -d ./CBE/

• Expected Results

Please find the expected log files and correction results at the folder D32_D33. The results for correcting D32 and D33 are presented under the folders of ABE and CBE, respectively.

Note: Noise2read may produce slightly different corrected result from these under D32_ABE and D33_CBE of D32_D33. This is because the easy-usable and automatic tuning of the classifiers’ parameters facilitates wide-range explorations, different best models are obtained for each training, but the final prediction results are stable within a certain range. We have discussed this in the Discussion section of our paper.

More examples for reproducing our experiments in this paper can be found at the Examples of the documentation

Feel free to contact me if you have any questions on running noise2read or are interested in noise2read.