Fig 1.
Illustration of observable and unobservable information.
Illustration of information contained in blood samples. The top row illustrated three (super-)infections. The bottom row illustrates the respective information about the infection that can be reconstructed from a blood sample. The first individual was infected by m = 4 lineages, three times with the orange and once with the blue lineage. Hence, the orange and blue lineages are observed in the blood sample, while the pink and green lineages were not observed. In the middle, a super-infection with m = 3 lineages is illustrated which differs from the first infection but results in the same observation. All four lineages were infecting in the third example, however, m = 6 super-infections occurred.
Fig 2.
The figure shows the relative bias in % of the BCMLE (solid lines) and MLE
(dashed lines) as a function of the true parameter ψ based on simulated data created by the conditional Poisson model. Each panel assumes a different lineage-frequency distribution p shown at the top of each panel. Each colored line corresponds to a different sample size N.
Fig 3.
Similar to Fig 2 but for the coefficient of variation in %. The dotted lines are the respective predictions based on the Cramér-Rao lower bounds.
Fig 4.
Robustness of MOI estimates against model violations.
The figure shows the bias of the BCMLE (solid lines) and MLE
(dashed lines) in % as a function of the true parameter
. The datasets are generated from the conditional negative binomial model whereas the estimates are derived from the conditional Poisson model. The panels in different rows correspond to different levels of overdispersion indicated by α. Panels on the left and right assume a different lineage-frequency distributions p shown at the top of each panel. Line colors correspond to a different sample sizes (N).
Fig 5.
Variance if MOI estimates under model violations.
Similar to Fig 4 but for the coefficient of variation in %.
Fig 6.
Bias of heuristically adjusted MOI estimators.
Shown is the relative bias in % of the heuristically adjusted estimators (HBCMLE1—, HBCMLE2—
, HBCMLE3—
) along with the relative bias in % of the MLE
and BCMLE
as a function of the true parameter ψ based on simulated data created by the conditional Poisson model. Each panel assumes a different lineage-frequency distribution p shown at the top of each panel. Colors correspond to different estimators. The relative bias in each panel is derived from S = 100, 000 randomly generated datasets of sample size 40.
Fig 7.
Variance of heuristically adjusted MOI estimators.
Similar to Fig 6 but for the coefficient of variation in %. The dotted lines are the respective predictions based on the Cramér-Rao lower bounds.