Modified README.md to update the citation of the (now published) paper

See merge request !1

Some scripts were missing (_Functions_MCMC.R and _Species_Fitting.R) to properly use the code. They're back.

Hélène Cecilia, Roosmarie Vriens, Jeroen Kortekaas, Paul J. Wichgers Schreur, Mariken de Wit, Raphaëlle Métras, Pauline Ezanno, Quirine A. ten Bosch (2021)
Heterogeneity of Rift Valley fever virus transmission potential across livestock hosts, quantified through a model-based analysis of host viral load and vector infection.
bioRxiv (https://doi.org/10.1101/2021.10.22.465395)

See README for more information on how to explore and use this code base.

Dose response relationship fitted to reviewed data, different for Aedes and Culex
Viremia converted to net infectiousness

Next : Species data

Model from Petrie et al. 2013 Reducing Uncertainty in Within-Host Parameter Estimates of Influenza Infection by Measuring Both Infectious and Total Viral Load
Package odin used to speed up the model (convert the ODE system in C)
MCMC works when fixing delta, p and kl and estimating beta, xi, dinf and ch

First try at optimizing sd for the mcmc sampling. Results saved in Output. Nice acceptance rates but not always converging in the right place. Extreme acceptance rate (0 and 1) for some parameters. 15 optimization rounds of MCMC with 1000 iterations each lasted 25mins.

Model modification to differenciate between the observation process (RNA measures only M segments) and the entirety of non infectious particles. Need to decide whether empty particles are counted as non infectious, depends if they play a role in our self-limited infection models

fix issue: do.call needs a list of parameters, each must have a name to be correctly passed to simulate_basic_model. Need to make sure that allpars is made of named vectors only.

Fixed file tree (hopefully)

Small code update : spotted anomaly when running plot.posterior.profiles : not peaking at the right place (contrary to plot.likelihood.profiles)

Many wrapper to allow generic functions to be applied systematically for a given model/dataset/parameters without passing it as parameter
Maybe it adds confusion to the code and is not necessary ?
Negative numbers returned by odesolver are set equal to 0
If MCMC sampling gives Inf for a parameter, it is automatically not accepted (to avoid likelihood = NA)
Regular message "DLSODA corrector convergence failed repeatedly" see if it's worrying and should be fixed
TODO : format results to allow the use of BayesianTools diagnostics plots

Code to test basic model (agreed on 21/09 meeting, see manuscript_draft "Target cell-limited model" )
Automatize plots of likelihood profiles to see if they peak at the right place (for known parameter values, simulated data)
First observation : basic model seems insufficient to reproduce the dynamics present in our data, particularly the decay
The simulated data is close to what can be observed for sheep that died during the experiment (before any decay)
TO DO: Plug Quirine's MCMC code and try to make it work

Script Rift_WithinHost_Temperature.R exploring correlations between viral loads and temperature data
Script Rift_WithinHost_NLOPTR_Testing.R to test a package for least-square fitting

.bib file to update as we read more

Temporary new directory "start_fresh" where I proceed to clean the code to start with a clean state.
Optimize_Mixing script finds the optimized sd for parameter sampling by fitting one parameter at a time.
The goal is then to see if these sd give good mixing once we try to fit several parameters.

Segment analysis : corrected confidence interval function (sample_from_CI) + compute CI of contribution of covariance (~ non randomness of packaging efficiency)
segmentNLL.nested function to test nested models
segmentNLL for the model with different p and covariances (the one retained for the analysis)

Segment analysis: likelihood ratio test to justify the use of species-specific parameters. Contribution of covariances to complete particles.

Segment analysis : going further on analysing the results, comparing fits of different datasets + quantifying the contribution of covariances

Updated segments scripts to convert covariance to correlation, and compute confidence intervals on the fits by drawing 1000 combinations of parameters (normal distribution).
Added a Functions script to have a lighter Main, following pattern agreed by QtB and HC.

Segment script modified to automatize the possibility to fix/estimate given parameters and perform likelihood ratio tests.
In the negative log-likelihood function, p_m and p_l are therefore defined as multiplicative coefficient to be applied to p_s, so that keeping p constant for the three segments is equivalent to having p_m and p_l = 1

- same probability p for all segments ("one p")
- different probabilities, no correlations ("three p")
- different probabilities, with correlations ("corr")

The models are compared with likelihood ratio tests (by hand).
This was done for the different kinds of data.
The most complex model performs significantly better for all data.

3 parameters, one for each pair of segments
If the two segments of the pair are absent or present, the covariance parameter is added
If one is present and the other absent, the covariance parameter is substracted
Improves the fit for SML particles

Added RITRA data and scripts to create fitR models with different likelihoods and test their performance on simulated data

- does not account for correlation yet