4 ms·

I feel like there should be a big effort for investigating health effects of chemicals before they even get tested in the lab, basically going back to first pri

I feel like there should be a big effort for investigating health effects of chemicals before they even get tested in the lab, basically going back to first principles. Couldn't we build an inventory of every damn molecule occuring naturally in human bodies, tagged by the subsystems they are associated with (blood circulation, cell cores, digestion etc.) and then run simulations that tell you for a new chemical what they can bond with and what exposure is safe? Even just a molecular bonding database that you can query like this could be very powerful, maybe full scale simulation isn't even needed to get a qualitative view on health effects. Am I naive about what this could do or the feasibility of the implementation? Is it being done since a long time and chemical enterprises just keep it for themselves?

adrianN 11y ago
I have only a very cursory understanding of computational chemistry, but I believe that finding out what molecules bind to the proteins in your body in what way is a hard problem. This is the reason why pharma corporations have to do lots of experiments with the actual stuff even though simulations exist; the necessary use of heuristics in these simulations make results unreliable. And even with all that testing it happens quite frequently that a new drug happens to have serious adverse effects that are only discovered during late stage clinical testing. I therefore don't think that it's feasible to determine the safety of all the chemicals we dump in the environment in this way.
quanticle 11y ago
Couldn't we build an inventory of every damn molecule occuring naturally in human bodies, tagged by the subsystems they are associated with (blood circulation, cell cores, digestion etc.) and then run simulations that tell you for a new chemical what they can bond with and what exposure is safe? In a word? No. Not at all. Modeling the behavior of individual proteins is a NP-hard problem [0]. Modeling the behavior of multiple proteins interacting is even worse. And modeling the behavior of thousands of proteins, under the influence of thousands of unregulated chemicals is a combinatorial explosion of NP hard problems. [0] http://www.piercelab.caltech.edu/assets/papers/proeng02.pdf http://www.piercelab.caltech.edu/assets/papers/proeng02.pdf
- m_mueller 11y ago
  Thanks! Alright then. Am I guessing right that a powerful quantum computer could find the solution in constant time though? It seems to me molecular bonding is just an energy minimization problem. Also, is it really necessary to look at everything in combination? You mean things like catalysts? Are typical chemical compounds really as complex as proteins? I think Computational Biology is lately often getting breakthroughs that were quite unexpected, see for example http://www.hpcwire.com/2015/01/16/nvidia-gpus-unfold-secrets-human-genome/ http://www.hpcwire.com/2015/01/16/nvidia-gpus-unfold-secrets....

adrianN 11y ago

I have only a very cursory understanding of computational chemistry, but I believe that finding out what molecules bind to the proteins in your body in what way is a hard problem. This is the reason why pharma corporations have to do lots of experiments with the actual stuff even though simulations exist; the necessary use of heuristics in these simulations make results unreliable. And even with all that testing it happens quite frequently that a new drug happens to have serious adverse effects that are only discovered during late stage clinical testing. I therefore don't think that it's feasible to determine the safety of all the chemicals we dump in the environment in this way.

quanticle 11y ago

Couldn't we build an inventory of every damn molecule occuring naturally in human bodies, tagged by the subsystems they are associated with (blood circulation, cell cores, digestion etc.) and then run simulations that tell you for a new chemical what they can bond with and what exposure is safe? In a word? No. Not at all. Modeling the behavior of individual proteins is a NP-hard problem [0]. Modeling the behavior of multiple proteins interacting is even worse. And modeling the behavior of thousands of proteins, under the influence of thousands of unregulated chemicals is a combinatorial explosion of NP hard problems. [0] http://www.piercelab.caltech.edu/assets/papers/proeng02.pdf http://www.piercelab.caltech.edu/assets/papers/proeng02.pdf

m_mueller 11y ago
Thanks! Alright then. Am I guessing right that a powerful quantum computer could find the solution in constant time though? It seems to me molecular bonding is just an energy minimization problem. Also, is it really necessary to look at everything in combination? You mean things like catalysts? Are typical chemical compounds really as complex as proteins? I think Computational Biology is lately often getting breakthroughs that were quite unexpected, see for example http://www.hpcwire.com/2015/01/16/nvidia-gpus-unfold-secrets-human-genome/ http://www.hpcwire.com/2015/01/16/nvidia-gpus-unfold-secrets....