Home
Welcome to MetaFX wiki page!
MetaFX (METAgenomic Feature eXtraction) is an open-source library for feature extraction from whole-genome metagenome sequencing data and classification of groups of samples.
The idea behind MetaFX is to introduce the feature extraction algorithm specific for metagenomics short reads data. It is capable of processing hundreds of samples 1-10 Gb each. The distinct property of suggest approach is the construction of meaningful features, which can not only be used to train classification model, but also can be further annotated and biologically interpreted.
To run MetaFX, one need to clone repo with all binaries.
git clone https://github.com/ctlab/metafx
cd metafxThen add MetaFX binary directory to path.
export PATH=/path/to/metafx/bin:$PATHFor permanent use, add the above line to your ~/.profile or ~/.bashrc file.
Requirements:
- JRE 1.8 or higher
- python3
- python libraries listed in
requirements.txtfile. Can be installed using pip
python -m pip install --upgrade pip
pip install -r requirements.txt-
coreutils required for macOS (e.g.
brew install coreutils) - If you want to use
metafx metaspadespipeline, you will also need SPAdes software. Please follow their installation instructions (not recommended for first-time use).
Scripts have been tested under Ubuntu 18.04 LTS, Ubuntu 20.04 LTS, macOS 11 Big Sur, and macOS 12 Monterey, and should generally work on Linux/macOS.
To run MetaFX use the following syntax:
metafx <pipeline> [<Launch options>] [<Input parameters>]To view the list of supported pipelines run metafx -h or metafx --help.
To view help for launch options and input parameters for selected pipeline run metafx <pipeline> -h or metafx <pipeline> --help.
MetaFX supports both single-end and paired-end input files. For correct detection of paired-end reads, files should be named with suffixes "_R1"&"_R2" or "_r1"&"_r2" after sample name before extension. For example, sample_r1.fastq&sample_r2.fastq, or reads_R1.fq.gz&reads_R2.fq.gz.
By running MetaFX a working directory is created (by default ./workDir/).
All intermediate files and final results are saved there.
MetaFX is a toolbox with a lot of modules divided into three groups:
There are pipelines aimed to extract features from metagenomic dataset without any prior knowledge about samples and their relations. Algorithms perform (pseudo-)assembly of samples separately and construct the de Bruijn graph common for all samples. Further, graph components are extracted as features and feature table is constructed.
Performs unsupervised feature extraction and distance estimation via MetaFast (https://github.com/ctlab/metafast/).
metafx metafast -t 2 -m 2G -w wd_metafast -k 21 -i test_data/test*.fastq.gz -b1 100 -b2 500Input parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filenames> | list of reads files from single environment. FASTQ, FASTA, gzip- or bzip2-compressed [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
| -l, --min-seq-len <int> | minimal sequence length to be added to a component (in nucleotides) [default: 100] |
| -b1, --min-comp-size <int> | minimum size of extracted components (features) in k-mers [default: 1000] |
| -b2, --max-comp-size <int> | maximum size of extracted components (features) in k-mers [default: 10000] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
For backward compatibility with supervised feature extraction, every sample belongs to category all.
| file | description |
|---|---|
| wd_metafast/categories_samples.tsv | tab-separated file with 3 columns: <category>\t<present_samples>\t<absent_samples> |
| wd_metafast/samples_categories.tsv | tab-separated file with 2 columns: <sample_name>\t<category> |
| wd_metafast/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
| wd_metafast/contigs_all/seq-builder-many/sequences/component.seq.fasta | contigs in FASTA format as features (suitable for annotation and biological interpretation) |
| wd_metafast/matrices/dist_matrix_<DATE_TIME>.txt | Distance matrix between every pair of samples built over feature table using Bray-Curtis dissimilarity |
| wd_metafast/matrices/dist_matrix_<DATE_TIME>_heatmap.png | Heatmap and dendrogram of samples based on distance matrix |
Example of the resulting heatmap with dendrogram:
Performs unsupervised feature extraction and distance estimation via metaSpades (https://cab.spbu.ru/software/meta-spades/)
metafx metaspades -t 2 -m 4G -w wd_metaspades -k 21 -i test_data/test_*.fastq.gz -b1 500 -b2 5000Input parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filenames> | list of PAIRED-END reads files from single environment. FASTQ, FASTA, gzip-compressed [mandatory] |
| --separate | if TRUE use every spades contig as a separate feature (-l, -b1, -b2 ignored) [default: False] |
| -l, --min-seq-len <int> | minimal sequence length to be added to a component (in nucleotides) [default: 100] |
| -b1, --min-comp-size <int> | minimum size of extracted components (features) in k-mers [default: 1000] |
| -b2, --max-comp-size <int> | maximum size of extracted components (features) in k-mers [default: 10000] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
For backward compatibility with supervised feature extraction, every sample belongs to category all.
| file | description |
|---|---|
| wd_metaspades/categories_samples.tsv | tab-separated file with 3 columns: <category>\t<present_samples>\t<absent_samples> |
| wd_metaspades/samples_categories.tsv | tab-separated file with 2 columns: <sample_name>\t<category> |
| wd_metaspades/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
| wd_metaspades/contigs_all/seq-builder-many/sequences/component.seq.fasta | contigs in FASTA format as features (suitable for annotation and biological interpretation) |
| wd_metaspades/matrices/dist_matrix_<DATE_TIME>*.txt | Distance matrix between every pair of samples built over feature table using Bray-Curtis dissimilarity |
| wd_metaspades/matrices/dist_matrix_<DATE_TIME>*_heatmap.png | Heatmap and dendrogram of samples based on distance matrix |
Example of the resulting heatmap with dendrogram:
There are pipelines aimed to extract group-relevant features based on metadata about samples such as diagnosis, treatment, biochemical results, etc. Dataset is split into groups of samples based on provided metadata information and group-specific features are constructed based on de Bruijn graphs. The resulting features are combined into feature table.
Performs supervised feature extraction using group-specific k-mers. K-mer is considered to be group-specific if it is present in at least G samples of certain group and absent in all other samples.
metafx unique -t 2 -m 4G -w wd_unique -k 21 -i test_data/sample_list_train.txtInput parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filename> | tab-separated file with 2 values in each row: <path_to_file>\t<category> [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
| --min-samples <int> | k-mer is considered group-specific if present in at least G samples of that group. G iterates in range [--min-samples; --max-samples] [default: 2] |
| --max-samples <int> | k-mer is considered group-specific if present in at least G samples of that group. G iterates in range [--min-samples; --max-samples] [default: #{samples in category}/2 + 1] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
| file | description |
|---|---|
| wd_unique/categories_samples.tsv | tab-separated file with 3 columns: <category>\t<present_samples>\t<absent_samples> |
| wd_unique/samples_categories.tsv | tab-separated file with 2 columns: <sample_name>\t<category> |
| wd_unique/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
| wd_unique/contigs_<category>/seq-builder-many/sequences/component.seq.fasta | contigs in FASTA format as features (suitable for annotation and biological interpretation) |
Performs supervised feature extraction using statistically significant k-mers. K-mer significance for a certain group is determined in two steps. Firstly, Pearson's chi-squared test for homogeneity filters out k-mers, which have the same distribution between categories. Secondly, for each pair of categories Mann–Whitney U test is applied to select k-mers with different occurrences between categories.
metafx stats -t 2 -m 4G -w wd_stats -k 21 -i test_data/sample_list.txt --pchi2 0.05 --pmw 0.05Input parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filename> | tab-separated file with 2 values in each row: <path_to_file>\t<category> [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
| --pchi2 <float> | p-value for chi-squared test [default: 0.05] |
| --pmw <float> | p-value for Mann–Whitney test [default: 0.05] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
| file | description |
|---|---|
| wd_stats/categories_samples.tsv | tab-separated file with 3 columns: <category>\t<present_samples>\t<absent_samples> |
| wd_stats/samples_categories.tsv | tab-separated file with 2 columns: <sample_name>\t<category> |
| wd_stats/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
| wd_stats/contigs_<category>/seq-builder-many/sequences/component.seq.fasta | contigs in FASTA format as features (suitable for annotation and biological interpretation) |
Performs supervised feature extraction using group-colored de Bruijn graph. All k-mers are colored based on their relative occurrences in categories and further de Bruijn graph is split into colored components.
Important! This module supports up to 3 categories of samples. If you have more, consider using other MetaFX modules.
metafx colored -t 2 -m 4G -w wd_colored -k 21 -i test_data/sample_list_3.txt --linear --n-comps 100 --perc 0.8Input parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filename> | tab-separated file with 2 values in each row: <path_to_file>\t<category> [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
| --total-coverage | if TRUE count k-mers occurrences in colored graph as total coverage in samples, otherwise as number of samples [default: False] |
| --separate | if TRUE use only color-specific k-mers in components (does not work in --linear mode) [default: False] |
| --linear | if TRUE extract only linear components choosing the best path on each graph fork [default: False] |
| --n-comps <int> | select not more than <int> components for each category [default: -1, means all components] |
| --perc <float> | relative abundance of k-mer in category to be considered color-specific [default: 0.9] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
| file | description |
|---|---|
| wd_colored/categories_samples.tsv | tab-separated file with 3 columns: <category>\t<present_samples>\t<absent_samples> |
| wd_colored/samples_categories.tsv | tab-separated file with 2 columns: <sample_name>\t<category> |
| wd_colored/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
| wd_colored/contigs_<category>/seq-builder-many/sequences/component.seq.fasta | contigs in FASTA format as features (suitable for annotation and biological interpretation) |
There are pipelines for analysis of the feature extraction results. Methods for samples similarity visualisation and training machine learning models are implemented. Classification models can be trained to predict samples' properties based on extracted features and to efficiently process new samples from the same environment.
Results of one of feature extraction MetaFX modules are required for further analysis
Performs PCA dimensionality reduction and visualisation of samples based on extracted features.
metafx pca -w wd_pca -f wd_stats/feature_table.tsv -i wd_stats/samples_categories.tsv --showInput parameters
| parameter | description |
|---|---|
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -f, --feature-table <filename> | file with feature table in tsv format: rows – features, columns – samples ("workDir/feature_table.tsv" can be used) [mandatory] |
| -i, --metadata-file <filename> | tab-separated file with 2 values in each row: <sample>\t<category> ("workDir/samples_categories.tsv" can be used) [optional, default: None] |
| --name <filename> | name of output image in workDir [optional, default: pca] |
| --show | if TRUE print samples' names on plot [optional, default: False] |
Output files
| file | description |
|---|---|
| wd_pca/pca.png | plot of 2-dimensional PCA in png format |
| wd_pca/pca.svg | plot of 2-dimensional PCA in svg format |
Example of the resulting PCA plot:
Executes Random Forest machine learning algorithm to train classification model based on extracted features.
metafx fit -w wd_fit -f wd_unique/feature_table.tsv -i wd_unique/samples_categories.tsvInput parameters
| parameter | description |
|---|---|
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -f, --feature-table <filename> | file with feature table in tsv format: rows – features, columns – samples ("workDir/feature_table.tsv" can be used) [mandatory] |
| -i, --metadata-file <filename> | tab-separated file with 2 values in each row: <sample>\t<category> ("workDir/samples_categories.tsv" can be used) [optional, default: None] |
| --name <filename> | name of output trained model in workDir [optional, default: rf_model] |
Output files
| file | description |
|---|---|
| wd_fit/rf_model.joblib | binary file with pre-trained classification model suitable for metafx predict module or further analysis |
This module processes new samples and counts feature values based on previously extracted features.
It can be used in the following case. A dataset with known samples categories have been processed, i.e. features were extracted and classification model was trained via fit or cv method.
Further new samples with unknown categories are needed to be studied.
Samples should be preprocessed with current calc_features method to get feature table, which can be used in predict module.
metafx calc_features -t 2 -m 4G -w wd_calc_features -k 21 -d wd_unique \
-i test_data/test_A_R1.fastq.gz test_data/test_A_R2.fastq.gz \
test_data/test_B_R1.fastq.gz test_data/test_B_R2.fastq.gz \
test_data/test_C_R1.fastq.gz test_data/test_C_R2.fastq.gz Input parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filenames> | list of reads files from single environment. FASTQ, FASTA, gzip- or bzip2-compressed [mandatory] |
| -d, --feature-dir <dirname> | directory containing folders with components.bin file for each category and categories_samples.tsv file. Usually, it is workDir/ from other MetaFX modules (unique, stats, colored, metafast, metaspades) [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
| --kmers-dir <dirname> | directory with pre-computed k-mers for samples in binary format [optional] |
Output files
| file | description |
|---|---|
| wd_calc_features/feature_table.tsv | tab-separated numeric features file: rows – features, columns – samples |
Executes Random Forest machine learning method to classify new samples based on pre-trained model.
Generally, --feature-table is obtained via MetaFX calc_features module and --model via MetaFX fit or cv module.
metafx predict -w wd_predict -f wd_calc_features/feature_table.tsv --model wd_fit/rf_model.joblibInput parameters
| parameter | description |
|---|---|
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -f, --feature-table <filename> | file with feature table in tsv format: rows – features, columns – samples ("workDir/feature_table.tsv" can be used) [mandatory] |
| --model <filename> | file with pre-trained classification model, obtained via fit or cv module ("workDir/rf_model.joblib" can be used) [mandatory] |
| -i, --metadata-file <filename> | tab-separated file with 2 values in each row: <sample>\t<category> to check accuracy of predictions [optional, default: None] |
| --name <filename> | name of output file with samples predicted labels in workDir [optional, default: predictions] |
Output files
| file | description |
|---|---|
| wd_predict/predictions.tsv | tab-separated file with 2 values in each row: <sample name>\t<predicted category> |
Example of the resulting file:
$ cat wd_predict/predictions.tsv
test_A A
test_C C
test_B BAs expected, all samples' categories were correctly predicted consistent with the file names.
Executes Random Forest machine learning algorithm to train classification model based on extracted features and check accuracy via cross-validation. Optionally, it can perform grid search of optimal parameters for classification model based on cross-validation.
metafx cv -t 2 -w wd_cv -f wd_stats/feature_table.tsv -i wd_stats/samples_categories.tsv -n 2 --gridInput parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: 1] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -f, --feature-table <filename> | file with feature table in tsv format: rows – features, columns – samples ("workDir/feature_table.tsv" can be used) [mandatory] |
| -i, --metadata-file <filename> | tab-separated file with 2 values in each row: <sample>\t<category> ("workDir/samples_categories.tsv" can be used) [optional, default: None] |
| -n, --n-splits <int> | number of folds in cross-validation, must be at least 2 [optional, default: 5] |
| --name <filename> | name of output trained model in workDir [optional, default: rf_model_cv] |
| --grid | if TRUE, perform grid search of optimal parameters for classification model [optional, default: False] |
Output files
| file | description |
|---|---|
| wd_cv/rf_model_cv.joblib | binary file with pre-trained classification model suitable for metafx predict module or further analysis |
Executes Random Forest machine learning algorithm to train classification model based on extracted features and immediately apply it to classify new samples.
It can be used in case when all samples are available at the beginning of the study, but only a part of them is labeled. And the goal is to classify the rest samples.
In that case --feature-table contains all samples, but --metadata-file contains information only about part of samples.
To model such a situation, we will use wd_colored/feature_table.tsv with 12 samples.
Also, we need to create special test_labels.tsv file with only 6 samples labeled.
$ echo -e "1\tA\n11\tA\n2\tB\n22\tB\n3\tC\n33\tC" > test_labels.tsv
$ cat test_labels.tsv
1 A
11 A
2 B
22 B
3 C
33 Cmetafx fit_predict -w wd_fit_predict -f wd_colored/feature_table.tsv -i test_labels.tsvInput parameters
| parameter | description |
|---|---|
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -f, --feature-table <filename> | file with feature table in tsv format: rows – features, columns – samples ("workDir/feature_table.tsv" can be used) [mandatory] |
| -i, --metadata-file <filename> | tab-separated file with 2 values in each row: <sample>\t<category> ("workDir/samples_categories.tsv" can be used) [optional, default: None] |
| --name <filename> | name of output trained model in workDir [optional, default: model] |
Output files
| file | description |
|---|---|
| wd_fit_predict/model.joblib | binary file with pre-trained classification model suitable for metafx predict module or further analysis |
| wd_fit_predict/model.tsv | tab-separated file with 2 values in each row: <sample name>\t<predicted category> |
Example of the resulting file:
$ cat wd_fit_predict/model.tsv
111 A
1111 A
222 B
333 C
3333 C
33333 CAs expected, all samples' categories were correctly predicted (true labels can be found in wd_colored/samples_categories.tsv).
There are modules to help to manipulate with data files and speed up MetaFX data processing.
This module processes reads files and extract k-mers for each sample separately.
It can reduce running time and RAM usage of other MetaFX modules, if the directory with extracted k-mers provided to other modules with --kmers-dir parameter.
metafx extract_kmers -t 2 -m 4G -w wd_extract_kmers -k 21 \
-i test_data/test_A_R1.fastq.gz test_data/test_A_R2.fastq.gz \
test_data/test_B_R1.fastq.gz test_data/test_B_R2.fastq.gz \
test_data/test_C_R1.fastq.gz test_data/test_C_R2.fastq.gzInput parameters
| parameter | description |
|---|---|
| -t, --threads <int> | number of threads to use [default: all] |
| -m, --memory <MEM> | memory to use (values with suffix: 1500M, 4G, etc.) [default: 90% of free RAM] |
| -w, --work-dir <dirname> | working directory [default: workDir/] |
| -k, --k <int> | k-mer size (in nucleotides, maximum value is 31) [mandatory] |
| -i, --reads <filenames> | list of reads files from single environment. FASTQ, FASTA, gzip- or bzip2-compressed [mandatory] |
| -b, --bad-frequency <int> | maximal frequency for a k-mer to be assumed erroneous [default: 1] |
Output files
| file | description |
|---|---|
| wd_extract_kmers/kmers/ | directory with k-mers in binary format for each sample |
| wd_extract_kmers/stats/ | directory with k-mers statistics for each sample |
Minimal example is provided in the README page.
More sophisticated example is presented at MetaFX tutorial page with links to input and output data, all commands and commentaries.