Introduction

StrainPycon is a Python 3 package that can be used to disambiguate multiple strains in mixed samples of DNA. Mathematically, StrainPycon can solve binary blind source separation problems and compute certain high-dimensional integrals involving binary variables. The connection between these mathematical concepts and strain identification is discussed in the following journal article:

L. Mustonen, X. Gao, A. Santana, R.M. Mitchell, Y. Vigfusson, and L. Ruthotto,
A Bayesian framework for molecular strain identification from mixed diagnostic samples,
Inverse Problems 34(10), 105009, 2018,
https://doi.org/10.1088/1361-6420/aad7cd

StrainPycon builds on the StrainRecon.jl package written in Julia: https://github.com/lruthotto/StrainRecon.jl

Motivation

As a motivating example, suppose you have a blood sample infected by multiple Plasmodium falciparum malaria parasites. Assuming you have done PCR on chosen SNP sites, the number of calls that differ from the reference genome are indicative of what proportion of the strains have mutated at that SNP. StrainPycon is an approach for identifying the strains in the sample through disambiguation (deconvolution) without requiring any prior knowledge about the sample or the parasite. The process can also help assess the multiplicity of infection in the sample, which can aid malaria surveillance efforts, for instance.

Citation

If you use StrainPycon in your project, please cite the journal article above.

Full documentation

Please refer to the full documentation of StrainPycon at: https://www.ymsir.com/strainpycon/

Requirements

StrainPycon was tested in the following environment:

• 64-bit Linux
• Python 3.6.5 with NumPy 1.14.3

Basic usage

Usually, the user only wants to access a few methods from the StrainRecon class:

import strainpycon
S = strainpycon.StrainRecon()

Let us generate synthetic measurement data with three strains and 24 SNP sites and solve the inverse problem:

(measurements, strains, freq) = S.random_data(24, 3)
(strains_recon, freq_recon) = S.compute(measurements, 3)

Here, strains_recon should equal strains and freq_recon should equal freq.

Next, let us draw another random measurement, now with Gaussian additive noise. We compute the misfit, or negative log-likelihood, when the number of strains in the reconstruction varies from one to seven. Moreover, we compute posterior statistics to quantify uncertainty:

gamma = 0.1 # standard deviation of Gaussian noise
(measurements, strains, freq) = S.random_data(18, 4, gamma=gamma)
misfits = S.misfits(measurements, range(1,8))
(strains_mean, freq_mean, strains_dev, freq_dev) = S.posterior_stats(measurements, 4, gamma)

A complete description of the methods and detailed examples can be found on: See https://www.ymsir.com/strainpycon/

Known issues

StrainPycon does not support multi-threading yet.

Contacts

Please direct questions to: Ymir Vigfusson, Emory University, ymir.vigfusson@emory.edu