Skip to content

Latest commit

 

History

History
277 lines (224 loc) · 20.1 KB

README.md

File metadata and controls

277 lines (224 loc) · 20.1 KB
Raw

TM-MSAligner

A Multi-Objective Multiple Sequence Alignment Tool for Transmembrane Proteins

TM-MSAligner is a multiobjective software tool to align multiple transmembrane protein sequences (TMPs) based on the M2Align software. The core of TM-MSAligner is a multi-objective evolutionary algorithm based combining features of the reference algorithms NSGA-II and SPEA2. TM-MSAAligner provides an encoding for representing MSA solutions and its corresponding mutation and crossover operators, and it is able of reducing the computing time by exploiting the computing capabilities of common multi-core CPU computers.

Requirements

To use TM-MSAligner the following software packages are required:

Downloading and compiling

To download TM-MSAligner just clone the Git repository hosted in GitHub:

git clone https://github.com/jMetal/TM-MSAligner.git

Once cloned, you can compile the software and generate a jar file with the following command:

mvn package

This sentence will generate a directory called target which will contain a file called tm-msaligner-1.0-jar-with-dependencies.jar

Generate the Transmembrane Topology Prediction using the DeepTMHMM tool

With the aim to predict the membrane topology of both alpha-helical and beta-barrel transmembrane proteins, we have used DeepTMHMM, a Deep Learning Model for Transmembrane Topology Prediction and Classification. This tool requires one or more proteins in FASTA format and generate the results with a file in .3line format.

The first line contains the Sequence Name, the second line the sequence and teh third line the topology prediction information. Each letter represents the type of topology: I: inside, O: outside and M: membrane.

For example:

>MSCL_ECOLI | TM
MSIIKEFREFAMRGNVVDLAVGVIIGAAFGKIVSSLVADIIMPPLGLLIGGIDFKQFAVTLRDAQGDIPAVVMHYGVFIQNVFDFLIVAFAIFMAIKLINKLNRKKEEPAAAPAPTKEEVLLTEIRDLLKEQNNRS
IIIIIIIIIIIIIIIIIIMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMOOOOOOOOOOOOOOOOOOOOOOOOOOOOMMMM

Generate list of pre-computed alignments (needed for generating the initial population)

To generate the Initial population of our algorithm, TM-MSAligner requires a list of pre-computed alignments. This list can be generated by the state-of-the-art MSA software tools. Assuming that the FASTA file containing the sequences to align is named "FASTA_FILE", the commands to run the different tools are:

ClustalW2

clustalw2 -infile=FASTA_FILE -outfile=FASTA_FILE_clu" -output=FASTA -ALIGN -QUIET -OUTORDER=input

Mafft

mafft --auto --inputorder --quiet FASTA_FILE > FASTA_FILE_mafft

T_Coffee

t_coffee FASTA_FILE -output fasta

Muscle

muscle -in FASTA_FILE -fastaout FASTA_FILE_muscle" -quiet

Kalign

kalign -c input -f fasta -q -i FASTA_FILE -o FASTA_FILE_kalign

Probcons

probcons FASTA_FILE > FASTA_FILE_probcons

Fsa

fsa --refinement 100 FASTA_FILE > FASTA_FILE_fsa

Runing TM-MSAligner

Running TM-MSAligner to align a custom TMPs sequences dataset

To execute the TM-MSAligner algorithm to align a custom dataset of sequences, just run this command:

java -cp target/tm-msaligner-1.0-jar-with-dependencies.jar org.tm_msaligner.TM_MSAligner dataDirectory problemName NumberOfEvaluations PopulationSize NumberOfCores typeObserver frequencyObserver

Where:

  • dataDirectory: The path of the data directory. This directory must contain two components. First, the file with the TMPs sequences and the topology prediction information, called problemName_predicted_topologies.3line (See how generate this file in above section). And second, the Pre-Computed alignments files in FASTA format (.fasta extension) performed by state-of-the-art of MSA software. These pre-computed alignments are used to generate the Initial population of the algorithm. The algorithm search files with extension .fasta and add to the list. See details how to generate them in the above section.
  • problemName: Name of the custom TMP Sequences dataset.
  • NumberOfEvaluations: Number of the maximun evaluations of the algorithm.
  • PopulationSize: Size of the population of the algorithm.
  • NumberOfCores: Number of cores to use for the parallel execution of the algorithm.
  • observerType: Type of the Observer added to the algorithm. TM-MSAligner includes three different observers to plot the performance of the algorithm, use the following values: 0 -> None, 1 -> FitnessWriteFileObserver, 2 -> FitnessPlotObserver and 3 -> FrontPlotTM_MSAObserve. See details of each observer in the section below.
  • frequencyObserver: The frequency of the Observer. Can't be greater than NumberOfEvaluations.

An example to align a custom TMPs dataset called msl stored in the folder data/custom_tests, with 25000 evaluations of the algorithm, a population size of 100, using 4 cores and plotting the ParetoFront every each 500 evaluations, you have to run this command:

java -cp target/tm-msaligner-1.0-jar-with-dependencies.jar org.tm_msaligner.TM_MSAligner resources/custom_tests/ msl 25000 100 4 3 500

The results are saved in the folder called as: dataDirectory/results_currentSystemTimeMilliseconds/. See more details about the results generated by our software in the next section.

Runing TM-MSAligner to align a TMPs instance from BAliBASE benchmark

To test the performance of our software, we have used some multiple sequence alignment (MSA) benchmarkings, one of them, is the reference 7 of BAliBASE (Benchmark Alignment dataBASE), this dataset is made of 435 alpha-helical TMPs classified into 8 distinct families named: 7tm, msl, ion, acr, nat, dtd, photo and ptga.

We have added the BAliBASE-ref7 dataset in our resources folder. In this folder we have added the pre-computed alignments (resources/precomputed_solutions/ref7 subfolder) performed by state-of-the-art of MSA software: ClustalW, KAlign, TM-TCoffee, Muscle, Mafft, T-Coffee. And the original sequences with the transmembrane topology information (resources/benchmarks/ref7) for each BAliBASE-ref7 instance. The files are named as: refName_predicted_topologies.3line.

To execute the algorithm to align BAliBASE-ref7 dataset, just run this command:

java -cp target/tm-msaligner-1.0-jar-with-dependencies.jar org.tm_msaligner.BAliBASETest balibaseInstanceName typeObserver frequencyObserver

Where:

  • balibaseInstanceName: the BAliBASE ref7 instance name.
  • observerType: Type of the Observer added to the algorithm. TM-MSAligner includes three different observers to plot the performance of the algorithm, use the following values: 0 -> None, 1 -> FitnessWriteFileObserver, 2 -> FitnessPlotObserver and 3 -> FrontPlotTM_MSAObserve. See details of each observer in the next section.
  • frequencyObserver: The frequency of the Observer. Can't be greater than NumberOfEvaluations.

An example to align the ptga instance and plotting the ParetoFront every each 500 evaluations, you have to run this command:

java -cp target/tm-msaligner-1.0-jar-with-dependencies.jar org.tm_msaligner.BAliBASETest ptga 3 500

Solving the BAliBASE-ref7 benchmarking, TM-MSAligner considers the following pathfolders:

  • benchmarkPath: Directory with the original BAliBASE-ref7 sequences without align and the predicted_topologies.3line file, the path is joined as follows 'resources/benchmarks/ref7/' + balibaseInstanceName + '/'
  • dataFile: The filename of the original TMP sequences of BAliBASE instance, 'benchmarkPath' + balibaseInstanceName + '_predicted_topologies.3line'
  • preComputedMSAPath: Directory with the PreAlignments, TM-MSAligner reads all .fasta extension files stored in this folder, the path is joined as follows 'resources/precomputed_solutions/ref7/' + balibaseInstanceName +'/'

The results are saved in the folder called as: resources/tests/ref7/balibaseInstanceName/results_currentSystemTimeMilliseconds/. See more details about the results generated by our software in the next section.

Running TM-MSAligner in Parallel

To speed-up the execution of TM-MSAligner, two parallel schemes are provided: synchronous and asynchronous. In the synchronous scheme, the evaluation of the population of the evolutionary algorithm is performed in parallel. It is selected by indicating the number of cores parameter when running the AsyncTM_MSAligner runner.

The asynchronous scheme is provided by the AsyncTM_MSAligner runner, whichs take the same parameters as TM_MSAligner. The working of the parallel scheme is based on generating as many solutions to be evaluated as the number of available cores and, whenever a solution is evaluated, the main loop of the evolutionary algorithm is executed to generate a new solution that is sent to the idle core to be evaluated.

Results

Final Pareto Front and Alignments

The software generates the FUN file with the Pareto Front points (solutions) called: FUN_problemName.tsv, each point represents one alignment, whose are illustrated with the MSA HTML viewer called MSABrowser. The topology is colored with different values to can visualize the regions, quickly and easily. The alignments are saved as: MSASol.html. Finally, the main html page with the plot of the Pareto Front and the Solutions viewer is generated as resultsMSA_problemName.html We illustrated some screenshots of the results, with the selection of the best alignments (points) in the Pareto Front:

Solution selected with the highest Segment Aligned score
Solution selected with the highest Sum-Of-Pair with Topology Prediction score
Pareto-Front
Pareto-Front

Visualization of the Alignments with the highest (image above) and lowest (image below) Sum-Of-Pair with Topology Prediction scores
Alignment Considering that SOPwTP score penalizes the insertion of Openning Gaps, even more inside transmembrane regions, the alignment with the highest SOPwTP score has fewer TM regions broken than the lowest one.

Pareto Front Observer

In this example, solving the ptga instance of the BAliBASE-ref7 instance, the software plot the evolution of the Pareto-Front generated by our algorithm TM-MSAligner, the next figures, illustrates the Pareto-Fronts during different number of evaluations.

Different Pareto-Fronts
1000 Evaluations
Pareto-Front at the evaluation 3500 		3500 Evaluations
Pareto-Front at the evaluation 3500
13000 Evaluations
Pareto-Front at the evaluation 3500 		25000 Evaluations
Pareto-Front at the evaluation 3500

Fitness evolution Observer

In this example, solving the msl instance of the BAliBASE-ref7 instance, the software generated the file BestScores_msl.tsv with the evolution of both fitness scores implemented in our algorithm TM-MSAligner, the results are illustrated in the following pictures:

Evolution of the Fitness Scores
Sum-Of_Pair with Topology Prediction Score
Evolution of the Fitness Score Sum-Of_Pair with Topology Prediction 		Aligned Segments Score
Evolution of the Fitness Score Aligned Segments

Finding algorithm configurations with irace

We explain in this section the steps needed to use TM-MSAligner in combination with irace, a tool for automatic algorithm configuration (NOTE: experience with irace is a pre-requisite for understanding this section). All the needed stuff is stored in the resources/irace folder.

The contents of the irace folder is the following:

  • data: a copy of the data folder of the project.
  • instances-list.txt: a text file containing the name of the instances used as training set as well as the name of the file containing their reference fronts. The file currently contains the msl instance:
msl --referenceFrontFileName msl.csv
  • irace_3.5.1.tar.gz: the irace package
  • parameters.txt: a text file containing the parameter space in the irace format (note that parameters populationSize and maximumNumberOfEvaluations are included in the first two lines of the file):
maximumNumberOfEvaluations               "--maximumNumberOfEvaluations "           o     (10000)  
populationSize                           "--populationSize "           o         (50)  
algorithmResult                          "--algorithmResult "                     c       (externalArchive, population)                      
populationSizeWithArchive                "--populationSizeWithArchive "           i       (10, 200)                      | algorithmResult %in% c("externalArchive")
externalArchive                          "--externalArchive "                     c       (crowdingDistanceArchive, hypervolumeArchive, unboundedArchive) | algorithmResult %in% c("externalArchive")
#
ranking                                  "--ranking "                             c       (dominanceRanking, strengthRanking)                     
#
densityEstimator                         "--densityEstimator "                    c       (crowdingDistance, knn)                            
kValueForKNN                             "--kValueForKNN "                        i       (1, 3)                         | densityEstimator %in% c("knn")
#
variation                                "--variation "                           c       (crossoverAndMutationVariation)                     
offspringPopulationSize                  "--offspringPopulationSize "             i       (1, 400)                       | variation %in% c("crossoverAndMutationVariation")
crossover                                "--crossover "                           c       (SPX)                          | variation %in% c("crossoverAndMutationVariation")
crossoverProbability                     "--crossoverProbability "                r       (0.0, 1.0)                     | crossover %in% c("SPX")
mutation                                 "--mutation "                            c       (insertRandomGap, mergeAdjuntedGapsGroups, shiftClosedGaps, splitANonGapsGroup) | variation %in% c("crossoverAndMutationVariation")
mutationProbabilityFactor                "--mutationProbabilityFactor "           r       (0.0, 2.0)                     | mutation %in% c("insertRandomGap","mergeAdjuntedGapsGroups","shiftClosedGaps","splitANonGapsGroup")
#
selection                                "--selection "                           c       (tournament, random)                               
selectionTournamentSize                  "--selectionTournamentSize "             i       (2, 10)                        | selection %in% c("tournament")
  • run.sh: script to run irace. Parameters such as the number of cores to use are included in this file.
  • scenario: folder that contains the scenario.txt file needed by irace. The files generated by irace will be stored here.
  • resources: folder containing the reference fronts (in CSV format) of the instances using for training.

The steps to run irace are the following:

  1. Execute mvn package in the root directory of the project. As a result, the file called TM-MSAligner-1.0-jar-with-dependencies.jar will be generated in the target directory. This file has to be copied to irace directory.
  2. Run irace:
./run.sh scenarios/scenario.txt 1
  1. The output of the irace execution will be stored in the directory scenarios/execdir-1. The best found configurations are written at the end of the file irace.stdout.txt. An example is shown next:
# Best configurations as commandlines (first number is the configuration ID; listed from best to worst according to the sum of ranks):
112 --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 149 --externalArchive hypervolumeArchive --ranking dominanceRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 201 --crossover SPX --crossoverProbability 0.8054 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.8994 --selection tournament --selectionTournamentSize 9
71  --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 69 --externalArchive hypervolumeArchive --ranking strengthRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 111 --crossover SPX --crossoverProbability 0.9687 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.8188 --selection tournament --selectionTournamentSize 7
95  --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 146 --externalArchive hypervolumeArchive --ranking dominanceRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 178 --crossover SPX --crossoverProbability 0.9186 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.9294 --selection tournament --selectionTournamentSize 6
129 --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 114 --externalArchive hypervolumeArchive --ranking strengthRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 139 --crossover SPX --crossoverProbability 0.9293 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.8636 --selection tournament --selectionTournamentSize 7
77  --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 149 --externalArchive hypervolumeArchive --ranking strengthRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 50 --crossover SPX --crossoverProbability 0.9582 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.7155 --selection tournament --selectionTournamentSize 5
  1. Select one of the configurations, remove the first element (the number is an id for irace that is not needed), and add at the beginning the parameters ProblemName, referenceFrontFileName, and randomGeneratorSeed. With the resulting string, just execute the following command from the project root directory:
java -cp target/TM-MSAligner-1.0-jar-with-dependencies.jar org.tm_msaligner.ConfigurableTM_MSAligner --problemName msl --referenceFrontFileName resources/referenceFronts/msl.csv --maximumNumberOfEvaluations 10000 --populationSize 50 --algorithmResult externalArchive --populationSizeWithArchive 149 --externalArchive hypervolumeArchive --ranking dominanceRanking --densityEstimator crowdingDistance --variation crossoverAndMutationVariation --offspringPopulationSize 201 --crossover SPX --crossoverProbability 0.8054 --mutation mergeAdjuntedGapsGroups --mutationProbabilityFactor 1.8994 --selection tournament --selectionTournamentSize 9

The output of TM-MSAligner will be stored in the files VAR.csv (the aligments) and FUN.csv (the values of the scores).