In this work, we developed a bacterial MALDI TOF spectra identification approach without using standard spectra library, but only protein sequences from UniProt database. Genetic algorithm with double cross-validation was performed to get the proteins encoded by the most informative genes as protein panels to be considered for bacterial identification. At genus level, the identification accuracy exceeded 80%. Although the performance is still behind to the performance of library search based approach, it can be used as a complementary method when standard library is not available. It can also be used in diverse application areas such as clinical microbiology, biodefense, food safety and environmental health.
Python 3.5 with numpy, pandas, and matplotlib
[Anaconda] (https://anaconda.org/anaconda/python) is recommended.
BacteriaMSLF uses peak lists for bacteria identification. A peak list is stored in a space-separated text (.txt) file including the m/z and intensity of all the peaks from a mass spectrum of a strain of bacteria with the first column as m/z and the second column as intensity, e.g.:
4025.85 14.9166578754281
4179.39 10.0793564012325
4195.16 100
4279.94 13.6100699949239
4296.99 13.759296917905
4339.2 16.2279091177302
...
9998.94 4.53344723468826
10103.51 25.6021474910836
10245.55 14.5594503134826
10457.14 9.18592847464415
10929.3 2.39529317439903
11115.92 10.3301379804275
11339.22 8.75138710039803
11702.38 2.53578789353994
-
European Consortium of Microbial Resources Centres (EM-baRC): [http://www.embarc.eu/deliverables/EMbaRC_D.JRA2.1.4_D15.33_MALDI-TOF_DB.pdf]
-
Robert Koch-Institute: [http://www.microbe-ms.com/]
-
The Public Health Agency of Sweden: [http://spectra.folkhalsomyndigheten.se/spectra/]
BacteriaMSLF uses proteomic sequences in UniProt[[http://www.uniprot.org/]
to match the MALDI-TOF spectra, e.g.:
>tr|A0A087INJ0|A0A087INJ0_VIBVL UDP-N-acetylmuramoyl-L-alanyl-D-glutamate--2,6-diaminopimelate ligase OS=Vibrio vulnificus GN=murE PE=3 SV=1
MRNTMNLTNLLAPWLDCPELADITVQSLELDSRQVKQGDTFVAIVGHVVDGRQYIEKAIE
QGANAIIAQSCQQYPSGLVRYQQNVVIVYLEKLDEKLSQLAGRLYQHPEMSLIGVTGTNG
KTTITQLIAQWLELAGQKAAVMGTTGNGFLNALQPAANTTGNAVEIQKTLADLQQQGAKA
TALEVSSHGLVQGRVKALQFAAGVFTNLSRDHLDYHGTMEAYAQAKMTLFTEHQCQHAII
NLDDEVGAQWFQELKQGVGVSLYPQDASVKALWASSVAYAESGITIEFEGCFGQGRLHAP
LIGEFNATNLLLALATLLALGVDKQALLGSAANLRPVLGRMELFQVNSKAKVVVDYAHTP
DALEKALQALRVHCTGHLWAIFGCGGDRDKGKRPMMAEIAERLADHVVLTDDNPRSEDPA
MIVQDMLAGLTRADSAVVEHDRFSALQYALDNAQADDIILLAGKGHEDYQVLKHQTVHYS
DRESAQQLLGISS
The following examples are running in the IPython environment.
Run [main.py] to load the functions.
%run main.py
Load the unknown peak lists.
sample_path = 'data/sample.txt'
sample = IdentifySpectra(sample_path)
sample.get_filterd_pattern(thr)weight.csv are obtained from traing, which is same as the the gene weights in the Figure 3.
my_gene = pd.read_csv('weight.csv',index_col=0)['mean']
model_data = gene_to_model(my_gene)Use model_data to identify this unknown spectra.
sample.answer(model_data, the_threshold)Output:
{'genera': 'bacillus',
'score': 1.9266267268100001,
'species': ('toyonensis', 'anthracis'),
'time': 2.3}The result of identification at genus level is Bacillus. At species level, it can not be distinguished between Bacillus toyonensis, Bacillus anthracis. The right identification Bacillus anthracis is in the group.
uniprot_path = ...
clean = os.listdir('uniprot_path')
cleaning_data(clean)training_set = ...
training_data(training_set,double_cross=10)modification of pyevolve.
Usage: The original gene weight table is:
rpmc 1
rpmj 0.947787695
rpmh 0.650960175
rpme 0.236613986
rpmg 0.452056135
rplx 0.39878139
rpmd 0.38311829
rpsp 0.268586678
hupa 0.225921218
hupb 0.230108406
Optimization:
import ga
import training
gene_weight_table = ...
parameters = ...
training_set = ...
my_gene = Gene(gene_weight_table)
my_alth = Algorithm(training.validation(), parameters)
my_gene.optimize(my_alth, 50)
print (my_gene.weight)The optimized gene weight table is:
rpmc 0.999341229
rpmj 0.931126682
rpmh 0.65297138
rpme 0.453734286
rpmg 0.451085325
rplx 0.414614517
rpmd 0.337521184
rpsp 0.328949972
hupa 0.328484009
hupb 0.327136496
package for identification of bacteria spectra
cleaning and transforming fasta file into csv file
plotting fuctions in the paper
training functions
Ding Cheng, Liang Qiao and Peter Horvatovich, "Toward Spectral Library-Free Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry Bacterial Identification", Submitted.
BacteriaMSLF is distributed under a BSD license. See the LICENSE file for details.
Please report any problems directly to the github issue tracker. Also, you can send feedback to liang_qiao AT fudan.edu.cn.