Modeling cell death

[ggsun]

Our simulation currently terminates only if it runs into an error, or if it has completed a designated number of simulations. Some of these errors are caused by bugs, but some of them reflect that the cell was unable to maintain growth/homeostasis in the given conditions. It would be useful if we could "handle these exceptions" and make the simulations alert us on why it's unable to continue and terminate in a clean way, and perhaps even activate a programmed death response. This would also be a prerequisite if we want to model gene knockouts and determine essentiality.

[jmason42]

Other "deaths" that I can think of:

• cytoplasmic volume not commensurate with cell surface area (osmotic pressure)
• extreme ionic/pH deviations
• excess accumulation of some metabolites, to the point where we expect nonspecific binding to hinder off-target enzymes
• "cells" without chromosomes

Solver/algorithm failures (whether those be fatal errors or just GIGO) aren't really cell death, but in something massive like a KO experiment we'd still like for those to lead to termination. For smaller experiments we may want to run out the non-fatal failures and see what happens.

[eagmon]

I wonder if we could use clustering of states as a way to determine cell death. For example -- if we have a set of gene knockouts that leads to death and a set that does not, we can apply those knockouts, track their trajectories through state space (state being the vector of counts), and then apply clustering analysis to separate the two sets. This clustering could potentially generalize to discern death states that weren't in the initial sets.

[1fish2]

FYI: If wholecell/sim/divide_cell.py does not have any full chromosomes at the end of the simulation duration, it sets _isDead = True on the daughter cells, which makes the run_incremental() method do nothing. (Frozen cells?)

[eagmon]

A non-trivial direction: we could pull together all components of the membrane (as I began to do in the spring) and see if we can determine a function that maps membrane composition to a permeability value. I have seen this done with lipid composition; the presence of proteins makes it more challenging. High permeability would be a good indicator of cell death.

[mialydefelice]

This seems like something good to add. An easier first pass might just be to separate out computational errors vs non-growth. Later you could separate out non-growth vs. cell death.