Integration of gene expression data using tINIT

The gene expression data was derived from {"type":"entrez-geo","attrs":{"text":"GSE147507","term_id":"147507"}}GSE147507. The triplicate data for mock infected NHBE cells and SARS-CoV-2 infected NHBE cells was considered. The information regarding the transcript isoform was obtained from Mammalian Transcriptomic Database (For Bronchial Epithelial Cells). The dataset was normalized to TPM (Transcripts per Million reads) using the transcript lengths specific for NHBE cells (S2 Dataset). Custom scripts were written in R for the same.

The latest version of tINIT (version 2.0) was used to integrate the gene expression data into HumanGEM model[25]. The steps for reconstruction has been described in S2 Fig. COBRA Toolbox v3.0 was used for all metabolic flux analysis in MATLAB 2017b. We used Gurobi optimizer with the tINIT algorithm. Briefly, the highly curated HumanGEM model in a closed form i.e. all exchange reactions closed was fed to the program “getINITModel2” of RAVEN toolbox in MATLAB 2017b [53] along with the TPM normalized gene expression data. A gene expression cutoff of 1 TPM was used for the algorithm as reported earlier[25]. The integer optimization algorithm underlying tINIT tries to minimize the incorporation of transcripts below this level. The task list for tINIT was prepared as reported earlier[25]. The task list for the creation of context specific model of NHBE cells infected with SARS-CoV-2 has additional reactions (S2 Dataset). Apart from synthesizing non-essential amino acids, conduct oxidative phosphorylation and other basic metabolic tasks, the model was expected to conduct specialized tasks necessary for the growth of the virus. These additional reactions were the tasks to de-novo synthesize NAG needed for the synthesis of viral biomass and produce viral biomass i.e. Cov2VBOF from the components in the media (S2 Dataset). IBM CPLEX 12.1 (IBM Academic License) was used for running the tINIT program.

The resulting context specific models was constrained using the uptake rates from media components as reported in several literature (S2 Dataset). Flux Balance Analysis using either the HumanGEM biomass reaction or Cov2VBOF was used to validate the doubling time of the NHBE cells and SARS-CoV-2 respectively. This was compared to the experimentally measured data[41] and a t-test was conducted to show no significant differences between the experimental and theoretical growth rate for SARS-CoV-2.

Henceforth, we will call the context specific model for normal NHBE cells and SARS-CoV-2 infected NHBE cells as iNHBE and iNHBECov2. For calculation of Hamming distance, we constructed a reaction Boolean vector for each model. Briefly, we constructed a Boolean vector of length 13417 (corresponding to all the reactions in the base HumanGEM model) for each model with 1s and 0s i.e. whether a reaction from the base model was incorporated into the context-specific model or not.

It was important check for the sensitivity of the specific growth rate prediction with respect to variations in coefficient (mmoles/gdw) or fractional composition of biomass precursors. We varied the coefficients of biomass precursors one at a time by ±10% and calculated the growth rate by flux balance analysis. The FBA was set-up on the modified SARS-CoV-2 biomass equations with coefficients increased by ±10% one at a time as follows:

Here, c* is the modified biomass coefficient vector such that c* = (c₁, c_2, c₃ …. c_n±0.1c_n). Here, the coefficient for c_n biomass precursor has been manipulated by ±10% whereas the coefficient of other biomass precursors remain the same.

IBM CPLEX solver was used to conduct the flux balance analysis. The context-specific model of NHBE infected with SARS-CoV-2 i.e. iNHBECov2 was considered for the analysis constrained with HAM media.