OTU-predictor 1.1.0

Uses a randomForest model to predict which OTUs are present in a microbiome

OTU_predictor

OTU_predictor uses a trained RandomForestClassifier ML model to predict 'real' OTU presence from ancient metagenomic samples (although it's use is not limited to ancient samples). The training dataset consists of 200 simulated populations generated through InSilicoSeq and deaminated using gargammel. Each population contains between 5 and 20 microbial species with know, variable abundance. OTU_predictor uses input files generated in centrifgure, specifically centrifugeReport.txt files.

Install package

OTU_predictor is currently running on python 3.11. It may run on earlier python versions also, but this has not been extensively tested. The easiest way to install OTU_predictor is using pip. Either of the following commands will do this:

pip install OTU-predictor

or

pip install git+https://github.com/DrATedder/OTU_predictor.git

Basic Usage

1. Converting your data

OTU_predictor works with centrifugeReport.txt files. Before you can run the model prediction step, some minor format teaks are required (see example output below). This can be done in the following way:

import OTU_predictor

centrifugeReport = "/path/to/your/file_centrifugeReport.txt"
OTU_predictor.convert_file(centrifugeReport)

If this step is successful, you will see a message similar to the one below:

'Data file /path/to/your/file_centrifugeReport_data.txt created'

The output data format should look something like this:

name	taxID	taxRank	genomeSize	numReads	numUniqueReads	abundance	genus	presence	sim_abundance
Bacteria	2	superkingdom	0	127	103	0.00026298841815572643	NA	0	0
Azorhizobium	6	genus	5369772	1	0	2.070774946108082e-06	Azorhizobium	0	0
Azorhizobium caulinodans	7	species	5369772	3	0	6.212324838324246e-06	Azorhizobium	0	0
Buchnera aphidicola	9	species	602805	3	1	6.212324838324246e-06	Buchnera	0	0
Cellulomonas gilvus	11	species	3526441	15	0	3.106162419162123e-05	Cellulomonas	0	0
Phenylobacterium	20	genus	4379231	1	0	2.070774946108082e-06	Phenylobacterium	0	0
Shewanella	22	genus	5140018	10	1	2.0707749461080822e-05	Shewanella	0	0
Shewanella putrefaciens	24	species	4749735	2	1	4.141549892216164e-06	Shewanella	0	0
Myxococcales	29	order	9638245	171	0	0.00035410251578448204	NA	0	0
Myxococcaceae	31	family	9636120	9	0	1.863697451497274e-05	NA	0	0
Myxococcus	32	genus	9487953	10	0	2.0707749461080822e-05	Myxococcus	0	0
Myxococcus xanthus	34	species	9139763	47	10	9.732642246707986e-05	Myxococcus	0	0
Myxococcus macrosporus	35	species	8973512	20	8	4.1415498922161644e-05	Myxococcus	0	0
Archangiaceae	39	family	10085598	11	0	2.2778524407188902e-05	NA	0	0
Stigmatella	40	genus	10260756	2	0	4.141549892216164e-06	Stigmatella	0	0
Stigmatella aurantiaca	41	species	10260756	1	0	2.070774946108082e-06	Stigmatella	0	0
Cystobacter	42	genus	0	1	0	2.070774946108082e-06	Cystobacter	0	0

Note. You will notice the addition of three new columns. these are variables used by the model during training, and while it is essential for them to be included for the file to be valid, they are not interpreted as part of the prediction step. It is also worth pointing out that the 'tab-delimitation' is replaced by 'comma-delimitation'.

2. Making predictions

The make_predictions() function uses your data file and the trained model which packages with this distribution. It is possible to create your own model if this is preferable though. The basic steps to make predictions are as follows:

converted_data = "/path/to/your/file_centrifugeReport_data.txt"

OTU_predictor.make_predictions(converted_data)

The output will be a list (of dictionaries) similar to the one shown below:

[{'Species': 'Neisseria mucosa', 'TaxID': 488, 'Prediction': 1, 'Certainty': 0.68},
{'Species': 'Streptococcus sanguinis', 'TaxID': 1305, 'Prediction': 1, 'Certainty': 0.72},
{'Species': 'Actinomyces sp. oral taxon 414', 'TaxID': 712122, 'Prediction': 1, 'Certainty': 0.97},
{'Species': 'Olsenella sp. oral taxon 807', 'TaxID': 712411, 'Prediction': 1, 'Certainty': 0.88},
{'Species': 'Anaerolineaceae bacterium oral taxon 439', 'TaxID': 1889813, 'Prediction': 1, 'Certainty': 0.87},
{'Species': 'Desulfobulbus oralis', 'TaxID': 1986146, 'Prediction': 1, 'Certainty': 0.84}]

Note. As you can see from the output list, OTU (species - although it can be at any taxonomic level determined by centrifuge) and taxID are given, along with a certainty score. These scores will be between 0 and 1, with higher scores indicating increased certainty. Prediction: 1 is OTU presence in the sample. The model also determines (but does not show) OTU absence (Prediction: 0).

Users should choose a certainty score that fits their experimental purpose.