So I figured that since I am attending the drug discovery conference at Northeastern University I might as well jot down a few thoughts about the talks.
Leroy Hood: Institute for Systems Biology (Seattle).
Lots of optimism, which is usually the case with systems biology talks; being able to distinguish "wellness" genes from "disease" genes for everybody in about ten years, being able to map all disease-related biomarkers from blood analysis etc. But there were some interesting tidbits:
- Noise - especially biological noise - cannot be handled by traditional machine learning approaches. Signal to noise ratio is very low especially when picking biomarkers.
- SysBio can help pharma pick targets (which it is increasingly getting worse at).
- Cost can be minimized in optimal cases; eg. FDA approved Herceptin specific for 20% of patients in only 40-patient sample (Genentech).
- Descriptive and graphical models can be enormously useful; in fact complexity often precludes mathematical modeling.
- Example of prions injected into mouse: expression of 33% genes changed. Biological noise can be “subtracted” by judiciously picking strains that get rid of one effect and preserve others.
My own take on systems biology has always been that, while it is likely to become quite significant at some point:
a. It's going to take longer than we think.
b. Separating signal from noise and honing in on the handful of approaches which will be robust and meaningful is going to give us a lot of grief. This will likely be Darwinian selection at its best.
Patricia Hurter (Vertex): Formulation
For people like me in discovery, formulation is a whole new world. Compaction, rolling, powder flow, force-response curves; engineers would feel right at home with these concepts, and in fact they do. And of course, you don’t talk about anything less than 25 kilograms.
Eric Olson (Vertex): Cystic Fibrosis
- Most common mutation is F508del (targets 88% of patients)
- Two potential drugs; potentiators (for restoring function) and correctors (for localizing protein from ER to membrane surface).
- However, only potentiators needed for G551D mutation (targets 4% of patients). Ivacaftor increases probability of channel being open; more beating cilia (nice video).
- Development challenges: little CF expertise, limited patient pop, no defined preclinical and regulatory path, outcomes for proof-of-concept and phase 3 not well established for mechanism-of-action.
I thought that the development of Vertex's CF drugs is a model example of charting out drug development in a novel, unexplored area.
Arun Ghosh (Purdue): Darunavir
From a medicinal chemistry standpoint this was probably the most impressive. Ghosh is one of the very few academic scientists to have a drug (Darunavir) named after them. He described the evolution of Darunavir from the key idea of targeting the backbone of HIV protease; the belief was that while side-chains are different between HIV mutants, the backbone stays constant and therefore compound binding to the backbone would be effective against resistant strains.
This idea turned out to be remarkably productive, and Ghosh described a series of studies that just kept on improving potencies against virtually any mutant HIV strain that the biologists threw at the compound. It was a medicinal chemist’s dream; there was a wealth of crystal structure data, compounds routinely turned out to have picomolar potencies, and almost every single modification that the chemists designed worked exactly as expected. Some of this success was of course good luck, but that’s something that’s usually a given in drug discovery. Darunavir and its analogs got fast-track FDA approval against HIV strains that had failed to respond against every other medication. Ghosh’s study was a powerful reminder that the right kind of design principal can lead to exceptional success, even against a target that's been beaten to death.
George Whitesides (Harvard): Challenges
Interesting talk by Whitesides. A pretty laid back speaker. The first half was a general rumination on the state of pharma and drug discovery ("the current model of capitalism is not working"; "the FDA has become unreasonable"; "if the best we can do in cancer is to invent a drug that gives someone 3 extra months with a lot of side effects, then we are doing something wrong").
The second half concerned his work on the hydrophobic effect. The papers deal with ligand binding to carbonic anhydrase. Basically he found out that the so-called entropic signature of the hydrophobic effect (an increase in entropy from release of bound water molecules) is more complicated.
A few notes:
- Designing drugs is hard because we are robust, multi multiplexed complex systems.
- Cost of healthcare in the US is ~17% of GDP: also, no correlation between health cost and quality, as evidenced by low standing of US.
- Quoted Anna Karenina’s happy and unhappy families; has something to do with drug development. Every successful drug has its success in common, unsuccessful drugs are unsuccessful in their own way.
- Pharmaceutical crisis has nothing to with per se with science, everything to do with costs.
Finally, he made an important point: biochemists have always done experiments in dilute phosphate buffer. Interior of cell is anything but.
Favorite quote, regarding the limitations of animal models: "Whatever else you may think of me, I am not a large, hairless mouse”
Leroy Hood: Institute for Systems Biology (Seattle).
Lots of optimism, which is usually the case with systems biology talks; being able to distinguish "wellness" genes from "disease" genes for everybody in about ten years, being able to map all disease-related biomarkers from blood analysis etc. But there were some interesting tidbits:
- Noise - especially biological noise - cannot be handled by traditional machine learning approaches. Signal to noise ratio is very low especially when picking biomarkers.
- SysBio can help pharma pick targets (which it is increasingly getting worse at).
- Cost can be minimized in optimal cases; eg. FDA approved Herceptin specific for 20% of patients in only 40-patient sample (Genentech).
- Descriptive and graphical models can be enormously useful; in fact complexity often precludes mathematical modeling.
- Example of prions injected into mouse: expression of 33% genes changed. Biological noise can be “subtracted” by judiciously picking strains that get rid of one effect and preserve others.
My own take on systems biology has always been that, while it is likely to become quite significant at some point:
a. It's going to take longer than we think.
b. Separating signal from noise and honing in on the handful of approaches which will be robust and meaningful is going to give us a lot of grief. This will likely be Darwinian selection at its best.
Patricia Hurter (Vertex): Formulation
Eric Olson (Vertex): Cystic Fibrosis
- Most common mutation is F508del (targets 88% of patients)
- Two potential drugs; potentiators (for restoring function) and correctors (for localizing protein from ER to membrane surface).
- However, only potentiators needed for G551D mutation (targets 4% of patients). Ivacaftor increases probability of channel being open; more beating cilia (nice video).
- Development challenges: little CF expertise, limited patient pop, no defined preclinical and regulatory path, outcomes for proof-of-concept and phase 3 not well established for mechanism-of-action.
I thought that the development of Vertex's CF drugs is a model example of charting out drug development in a novel, unexplored area.
George Whitesides (Harvard): Challenges
Interesting talk by Whitesides. A pretty laid back speaker. The first half was a general rumination on the state of pharma and drug discovery ("the current model of capitalism is not working"; "the FDA has become unreasonable"; "if the best we can do in cancer is to invent a drug that gives someone 3 extra months with a lot of side effects, then we are doing something wrong").
The second half concerned his work on the hydrophobic effect. The papers deal with ligand binding to carbonic anhydrase. Basically he found out that the so-called entropic signature of the hydrophobic effect (an increase in entropy from release of bound water molecules) is more complicated.
A few notes:
- Designing drugs is hard because we are robust, multi multiplexed complex systems.
- Cost of healthcare in the US is ~17% of GDP: also, no correlation between health cost and quality, as evidenced by low standing of US.
- Quoted Anna Karenina’s happy and unhappy families; has something to do with drug development. Every successful drug has its success in common, unsuccessful drugs are unsuccessful in their own way.
- Pharmaceutical crisis has nothing to with per se with science, everything to do with costs.
Finally, he made an important point: biochemists have always done experiments in dilute phosphate buffer. Interior of cell is anything but.
Favorite quote, regarding the limitations of animal models: "Whatever else you may think of me, I am not a large, hairless mouse”
Regular follower of the blog here. I attended most of the talks throughout the day of the symposium and frankly the Whitesides talk was on a different frequency in comparison to the others. His thought process and understanding of the basic problems in drug discovery is just mind-blowing.
ReplyDeleteAlso, being a Northeastern grad student, it was a tad embarassing when not many people believe they had taken a "good thermodynamics course". I myself am one of culprits who having taken a course during my undergrad days, though just not confident about my understanding of its basic principles. But I agree (and you seem to as well, from your following post) with his observation that a good thermodynamics course is vital for drug discovery research.
~Krishna
Thanks for reading. Yes, GW's talk certainly was quite thought-provoking. I thought it was a little too scattered all over the place, but keynote speakers and especially generalists like GW usually are.
Delete"Lots of optimism, which is usually the case with systems biology talks...."
ReplyDeleteI think that's pretty much required in such presentations. I've seen some really neat "proof of principle" types of work come from this corner of the scientific community, but it's usually involved systems where a lot of work was already done to serve as a "testbed" of sorts.
"...if the best we can do in cancer is to invent a drug that gives someone 3 extra months with a lot of side effects, then we are doing something wrong..."
I understand that this was a drug discovery conference, so this is a perfectly reasonable sentiment. But I'm always reminded of my time at a cancer institute where a number of the clinicians said basically the following to me, "Figuring out better diagnostics and treatments to inhibit metastasis would probably save more lives than anything else."
"Finally, he made an important point: biochemists have always done experiments in dilute phosphate buffer. Interior of cell is anything but."
I think people are trying to make a move towards more physiologically sensible conditions - higher ionic strengths, mimicking molecular crowding as much as possible, and so on. It's just difficult - as I've said before to people, the tricky thing in biochemistry isn't being able to vary one parameter. It's to keep all of the others from varying.
True. My main complaint is that a lot of the sysbio work that I hear doesn't seem to be connected to rigorous molecular-level follow-up experiments. It's often global, network-level, broad speculation. I am not saying that the methodology is not rigorous, it's just that ultimately whether it works or not is still going to depend on the details of protein-ligand interactions (thermodynamics, anyone?).
DeleteAs for your point regarding physiologically relevant concentrations, I feel hearted by some of the more recent simulation work done with protein folding in crowded environments. Long way to go, but good start. I think Peter Kenney who comments here often has mentioned that we still lack a sound method for measuring drug concentrations inside cells and at target locations, something that I think bears directly on this issue of crowding.
There is fundamental problems with systems biology as described above. Namely it does not address the system at the systems level. A capability we would argue is critical for the advancement of medicine.
ReplyDeleteThe signal to noise ratio referred to presumably relates to finding which aspects of genetic or protein information are important for a given purpose. The approach is presumably to mine such data for correlations and trends that correspond to a particular disease or function and then experiment to confirm or deny the hypothesis.
System biology as they are describing is therefore an extension to data mining and/or knitting together pathway biology. A logical approach but sadly one doomed to be very slow, expensive and statistically extremely unlikely to produce a result. What it does not do is address biology at the systems level.
This may appear semantics but it disguises a fundamental scientific point that represents a bit of a crossroads for drug discovery. If you want to create therapeutics at the systems level then the approaches taken must be able to interpret how these systems function at the systems level.
Complex systems science has demonstrated that this prod and observe approach is statistically very very unlikely to work. Evolution has made biology complex with feedback loops, compensatory pathways and biological systems that can alter their wiring quickly with change of environmental factors. The use of network pharmacology tools can identify at the systems level the proteins etc that count for the function at hand and which ones are synergistic. This focuses the effort, accelerating it, making it affordable and perhaps most fundamentally of all converting the probability from near certainty of failure to high chance of success.