Supplemental data S3 . Persistent Mycobacterium tuberculosis infection in BALB/c mice.

1. (A) (B) Supplemental data S3. Persistent Mycobacterium tuberculosis infection in BALB/c mice. In mice, low dose aerosol infection with pools of transposon mutants of M. tuberculosis CDC 1551 established a stable lung bacterial burden (A) and mice survived with similar weight gain over the next 28 weeks with an increase in lung and spleen weights (B). In order to account for the physiological increase in organ weights in aging animals, organ weights were normalized using the following formula: sacrificed animal organ weight on day of sacrifice × (mean body weight on day after infection/sacrificed animal body weight on day of sacrifice).

2. The features were mapped to a single prediction of mutant attenuation through logistic regression, Pr(att) = exp(βx)/[1+exp(βx)], where βx is the scalar dot product of a vector of regression coefficients β with the feature vector x. Penalized regression with cross-validation was used to select a parsimonious, robust subset of 11 features. Mutants corresponding to the top predictions were selected for experimental tests. Networks for Mtb were constructed from genes (represented as vertices) and pairwise relationships (represented as edges). Different networks were constructed based on metabolism, physical interactions, and inferred functional associations. Mutants for previously tested genes were identified as attenuated (green) or having a null phenotype (red). Phenotypes were treated as probabilities that diffused throughout the network to untested genes. Probabilities diffused through different networks and were generated from additional phenotypes that may be informative for persistence (3rd diagram). Phenotypes were also used directly, without diffusion (4th diagram). Each gene had 28 possible features in all, denoted as a 28-element vector x for each gene. Supplemental data S3C. Illustration of predictive methods