Ramblings: January 2024

10 March 2024

Below are collected Twitter ramblings for January 2024, with the usual commentary interspersed.

This month’s photo of the month is looking out over Lake Lillinonah from the Upper Paugussett State Forest in Newtown, Connecticut. It’s a typical gray and dreary winter day, but also one of exceptional solitude on the trails. Quiet hikes like this in the dead of winter often offer deep opportunities for self-reflection.

If you’re of the sort that makes New Year’s resolutions, and are at all involved in the world of drug discovery, might I suggest for 2024: resolve to become a better pharmacologist. Pharmacology is like the Force of drug discovery: it penetrates and binds together everything that we do. In vivo experiments are far too often a black box to members of a discovery team, who rely on experts to read the tea leaves for them. Commit to becoming enough of an expert that you can understand the contours of this kind of data on your own. The free drug hypothesis is a great place to begin.

For the record, the correct response was “request for peer review”. I had blown through all three of these within the first 10 days of 2024. The predatory journal invite was the last piece to fall into place, but only because my spam filter provided some inbox protection. We’re headed into the heart of conference season as I’m writing this, and the new spam this year is predatory selling of conference attendee lists and their contact information. Maximum sleaze.

Your career isn’t a zero-sum game with everyone else’s. That’s not always obvious when you first come to industry either. I’ve said many times before that drug discovery is a team sport, and that goes for your career management too. Ultimately, even if your goal is to move up to a senior position in the organization, your ability to do so is going to depend a lot more on your people management and leadership skills than your scientific skills. So it’s good to cultivate the habit of lifting everyone up from the beginning. In my experience, scientists are often a little short on those soft skills that they’re going to need in the long term.

The Rule of 5 (Ro5) is very much on my mind these days because I work with PROTAC compounds that routinely break those rules. If the Ro5 was really a hard and fast rule, you’d never even try to make orally bioavailable compounds like PROTACs in the first place — and you’d have been wrong to not try. I was happy to publish a paper last year sharing some of what we’ve learned about how things behave in the beyond Rule of 5 (bRo5) space. I think we still have a lot more to learn about what makes molecules orally absorbed, and some of what Darryl posted above begins to speak to that.

One of the downsides of being a little further up the food chain at work these days is that some days, the only thing I wrestle to the ground is a good PowerPoint deck. But I still have a project or two to run, and if I get a rare full day to do nothing but project work, it’s awesome.

Maybe this goes hand-in-hand with the prior point. Now that I have a lot more demands on my time, I’m grateful when people take the time to communicate succinctly. It takes good active listening skills to really hear the question you were asked and then center yourself enough to answer it. When I hear people answering a different question to the one I asked, I wonder first if I could’ve asked it differently.

But experience has shown that more frequently folks are following some internal narrative on the question they wished to answer instead — maybe because they’re uncomfortable with, or unprepared to answer, the actual question. It’s those moments, especially for team leaders, when you need to focus hard. Every team becomes myopic to the team’s issues with time, and sometimes questions from outside the team, while uncomfortable, are just what’s needed to shake the tree. Take tough questions as a gift. And it’s okay to say you don’t know and need some time to think about it! It’s part of the job of senior leaders to challenge the team’s thinking. They’re not challenging you personally — or at least they shouldn’t be.

Only half-joking about this. There’s a lot of shoddy experimental practice in the doctoral dissertations of yesteryear that I guarantee AI tools will uncover in coming years.

People, can we just not? It’s 2024 for crying out loud. By now we should all have had enough experience with email to be a little more careful. Heck, most of our early career scientists have never known a world without email. Mine will be the last generation that didn’t grow up with it. Honestly, we’re already reaching the point where email is becoming passé to our most recent generations. I wonder what will be next to replace email?

Much more than when I started in industry 2+ decades ago, there are now more viable paths for a chemist to walk in their undergraduate and graduate training to gain a job in industry. In my day, you pretty much had to come from a total synthesis or synthetic organic methodology lab to have a fighting chance for a med chem or process job in pharma or biotech. And that training is good training! It gives you a great grounding in not just how to make organic molecules (which is bread-and-butter stuff in pharma), but also in a problem-solving mindset that translates well to industry jobs.

Today, though, things are changing. Now we hire a lot more people not from that background. They’re still from academic disciplines like medicinal chemistry or chemical biology that are close to, and partially overlapping with, the traditional organic synthesis background. While those folks may not have quite the same synthetic chops, I think there’s more appreciation today for the value of building a department with diversified skills and backgrounds. We (meaning I) work on some pretty out there molecules these days that would’ve gotten you laughed out of the room in an industrial med chem group 15-20 years ago. And that revolution for PROTACs was led from the chemical biology side. So why not hire some of them?

At present, there is no escaping this hard truth: clinical trials are going to mow down 90-95% of what we in research come up with as clinical candidates, even after putting our best foot forward. Simply speeding up research processes does nothing to directly address that failure rate, and thus nothing to address the overall efficiency and cost of moving things to market. You’re just lining more stuff up to fail at the same rate.

Which is why we should be skeptical of any new research thesis that cannot answer the question: how will this technology help to reduce the clinical failure rate? VCs could probably save their firms a shit ton of money by just holding fast on this one question and demanding a good answer from would-be founders. And saying no to the folks who want to accelerate the non-rate determining parts.

If you’re lucky, from time to time, someone comes along who changes your career arc. I had great high school chemistry teachers who put me on the path to a chemistry major. I loved organic chemistry as an undergraduate, which partially set my path for grad school. As a senior, I took this graduate level class on organometallic chemistry, and it changed my path again. At the beginning, I struggled with organometallic complexes. They behaved by some weird rules and never saw an octet that they liked. Chad Tolman had a way of cutting through the bullshit though and showed us how this chemistry was being used at scale in some heavy-duty industrial processes. He also showed us how they came up with the ligand cone angles in the pre-computational chemistry era by simply building models and then mashing them into the smallest possible volume — by hand! Sometimes simple is best.

By the end, I was friends with the d block and had a good grasp of fundamental processes like oxidative addition, reductive elimination, transmetalation, and the like. And more importantly, I knew I needed to find a way to work this chemistry into my graduate work. I joined Jim Leighton’s group at Columbia in large part because it was right at the intersection of organic synthesis and organometallic chemistry that I wanted.

This ties in with the earlier post about hitting on clinical trials as the rate-determining step. There’s so much tech out there these days that isn’t hitting on clinical trial failure. AI has been a favorite punching bag for me in this regard, but in reality it’s just the latest in a fairly long string of new ideas that suffer from a similar flaw. We don’t need to discover clinical candidates faster — especially derivative ones against well-trodden targets. What we do need to do on the research side, which the post above addresses, is to get better at picking the right target(s) to work on in the first place. The question remains, and is the burden of new entrants to answer: how will this technology help to reduce the clinical failure rate? An acceptable answer to that question could be that the new technology will allow us to address targets that we know are good but didn’t previously know how to drug. Or how to address a common drug resistance mechanism in cancer. Or something — anything. Just not throwing more clinical candidates out that are gonna get harrowed to the ground the same as they do now.

Which brings us to the Jensen Huang JPM24 story. I view AI tech in drug discovery with cautious optimism. To be honest, I’m not sure we’ve found all the right niches for it yet. I do think it’s already adding value in parts of the discovery process. Just not yet those parts of the process, alluded to in the posts above, that can compellingly answer the “how will this technology help to reduce the clinical failure rate?” question. And God bless Jensen, he goes right for the throat and says f*ck it, we’re just gonna hook up a big enough pile of GPUs until we can simulate biology. While I think that’s a fantastical pipe dream today, if we could ever do that, it would certainly go right to the heart of clinical trial failure. We’d have much better odds on picking targets, or pathways, or even phenotypes to pursue. Unfortunately, until we have a training set to input that isn’t garbage, I think Huang’s grand dream will stay just that: a dream. When I ask proponents of this kind of approach what the specific model input data will look like, it’s crickets. More on that in a minute.

There’s so much texture hiding in dose response curves, if only we’d take the time to look at them. Med chemists, don’t settle for just an IC50, or other summarized parameters — ask to see the curves! There’s an entire tweetorial waiting to be written someday on the finer points of interpreting dose response curves. Like what to do when you get a partial curve where the top or bottom are missing, when the Hill slope isn’t 1, understanding if the xC50 point is the inflection point or 50% response absolute, what defines the top and bottom of the scale, and the like. It’s a powerful intuitive sense that you can build with time and practice.

I sometimes call this behavior among project leads “Afraid to Succeed Syndrome.” You’ve got something good on your hands, but you hesitate: what if this, what if that? This business is already hard enough without project leads going through mental gymnastics to talk themselves out of progressing a compound. Maybe there’s an overhanging fear that progression will be expensive or there will be repercussions for pushing forward and then having the compound fail out on some advanced assay. But remember, the cost of not progressing a winner is almost always going to be higher. If a compound clears the gates in the screening cascade, keep pushing it! If you think the compound is going to fail for some reason, that might be cause to go back and interrogate whether or not the screening cascade, and associated progression gates, are sound. Sometimes there’s something unwritten holding you back, and it’s usually best to articulate whatever that is.

Everyone had a good chuckle at this. Never waste a good Texas carbon moment!

This was the last chapter in the Jensen Huang saga, at least for this month. The argument here is: let’s say we build some massive data set and successfully reduce biology to its parts. Starting from the parts, can we reassemble the whole? Nobel Laureate Philip Anderson argued over 50 years ago in his famous essay “More is Different” that such attempts at reduction were sure to be thwarted by emergent new rules governing the system at each new level of complexity. In the same way that it argues chemistry cannot be reduced to physics, biology cannot be reduced to chemistry.

Even if you wanted to try, proponents still need to answer the critical question regarding the dataset. If you don’t have a great dataset to begin the reduction effort from, it’s almost certain to fail. Every time someone points to the success of AlphaFold in this regard, I point to the decades of meticulously curated data in the PDB that made AlphaFold possible. No equivalent dataset exists for “biology” — nor is it likely to exist anytime in the next several decades, if ever. We would first need a revolution in the way we gather and curate data regarding biological systems, which is notoriously noisy and error-prone.

I grew up in a middle class suburban town that was >90% white. Being involved in the scientific community has exposed me to people, cultures, and thinking (and food! glorious food) from all over the world, and that’s shaped me (I think) into a more well-rounded adult. (Come for the science, stay for the culture?) The kid from the ‘burbs might never have escaped from being surrounded by a bunch of people who looked and thought just like him. It’s a journey that I highly recommend for anyone who still has the time and energy to take it. And I suspect would defuse a lot of the clown show that American politics has become in the last decade.

While I’m a big believer in rapid iteration in preclinical drug discovery, eventually in the clinic that mentality has to stop. Rapid iteration in patients has a way of both muddling determination of therapeutic benefit and can also be downright dangerous. What happened with thalidomide in Europe, and how close it came to happening in the US, isn’t that far in the rear-view mirror. And that says nothing of the pre-FDA era of patent medicines hawked directly to consumers by quacks. Our society rightly places a huge premium on patient safety, the tolerance for pharma making a safety error is low, and there’s a huge regulatory machine in place to make sure it stays that way. Undoubtedly we could have rapid iteration in the clinic that leads to better therapeutics faster, but it would come at a human cost that nobody in their right mind wants to pay.

Can anyone even imagine what it would be like if Apple was legally mandated to run large population safety studies on physical and emotional well-being before they released a new iPhone? And then show that the new iPhone was so much better than the old iPhone that it was worth releasing? And that a government agency would have final say on the release? This is a reality of pharma that’s completely foreign to Silicon Valley.

I feel the same, doggo. Embrace the free drug hypothesis.

That old saying that “if you have a question, it’s likely that others have the same question — but are just afraid to ask” is largely true. Never be too proud to admit your ignorance. I’ve lost count of the number of times I’ve prefaced a question with “This may be stupid, but…” or “This is coming from my own ignorance, but…” or “I’m new to this project, so I’m probably not as up on this topic as I should be, but…” It’s important for leaders in an organization to normalize this kind of behavior. Sometimes your question may indeed be a naïve question that everyone else is already up to speed on. If so: no big deal, now you’re up to speed too! But I’ve also seen many occasions where someone asks such a question from outside a team and hits on a sensitive topic that a team may have been skulking around, afraid to confront head-on — even though it may be a critical issue for the project.

I don’t know why in certain circles the myth of “the work speaks for itself” continues to promulgate. The work most certainly does not speak for itself until you give it a voice. Particularly in the context of making a hiring decision, industry folks are far less concerned with the specifics of what you did for your PhD (or other degree) anyway. What they really want to know is: that person did some cool work and solved some tough problems — what are the chances they’ll come into our company and be able to repeat that feat, again and again, on a different set of problems? We’re hiring that skill set in you, the individual, not the work you did.

Understand this truth, and it will set you free.