Early attempts to recreate (micro-)life in silico have begun around 15 years ago. A useful model would suggest a hypothesis that forces the model builder to do an experiment. Take the early effort of Drew Endy of University of California at Berkeley and John Yin of University of Wisconsin-Madison as a landmark. Their computational model incorporates everything we know about the way the T7 bacteriophage virus infects the infamous Escherichia coli. The seemingly impressive simulation includes how all T7's 56 genes translates into 59 proteins, how those subverts the host cell, and even how the viruses would evolve resistance to various RNA-based drugs. Despite including measures from near two decades of experiments, early models fail miserably in that there are still a huge number of degrees of freedom so that they can be tweaked to produce almost any behavior. These models are then just sketchy caricatures based on the traditional gene => RNA => protein basic sequence.
The billions of dollars initially invested in technologies such as sequencing, combinatorial chemistry and robotics haven't paid off as hoped because of the naive idea that you can redirect the cell in a desired way just by sending in a drug that inhibits only one protein. Indeed, you could draw a map of all the components of the simplest single-celled microorganism and put all the connecting arrows and still have absolutely no ability to predict anything.
Since around 10 years ago, some more mathematically-minded biologists have been putting forth an effort to use computer simulations to search for some unifying principles that could order the facts, rather than search for a pretentious single model. I.e., a purely reductionist, top-down approach to simulating cells. For instance, from the currently more sophisticated cell simulations one can argue that robustness is a good candidate to be one of these conjectured emerging universal properties. Knowingly, to survive and prosper (i.e., (self-)reproduce), cells must have backup systems and biological networks that tolerate interferences such as dramatic temperature swings, food supply changes, and toxic chemicals assaults. In this all-important context, virtual experiments run with the japanese E-Cell model, a single-celled "microbe" mostly built from genes borrowed from Mycoplasma genitalium - the smallest genome yet discovered in a self-reproducing life-form - indicate that even with a drastic change in the magnitudes of various genes expressions a cell's behavior can remain practically unchanged. Experiments in which the researcher adjust virtual cells to reflect the activity of a specific drug are revealing that the resulting dramatic changes in cellular state can lead to a very little efficacy on the underlying disease condition.
It has been vividly argued that what most strongly affects how a cell behaves in response to a drug or disease is not any manipulation of a particular gene or protein, but how all the genes and proteins interact dynamically - i.e., the story emerges from the links, which shift over time. As we know hardly anything about most biochemical systems, some modellers are taking an engineering approach by figuring out the basic laws the cell's behavior must obey. Perhaps the most famous example of this approach, pioneered by computer science demigod John R. Koza, is that of a set of programs genetically evolving to match entire actual reaction networks. As measured data on how cells process chemicals over time are piling up, this evolutionary approach could one day be used even to deduce the convoluted paths by which cells turn food into energy, growth and waste.
Other modellers are mathematically reconstructing biochemical networks from first-principles, subjecting them to required mass, electrical and thermodynamical constraints, and then predicting optimal solutions within the remaining (physically viable) ones. For example, a research group at the University of California at San Diego has predicted that Escherichia coli is optimized for growth, not energy production.
The observation that many biochemical problems most likely have an optimal answer has led some modellers to predict a near future with quantitative models of cell function, organ function and eventually whole-animal function, perfect drug discovery engines.