Some Thoughts on Biotech vs Pharma for Computational Chemists




A recent editorial by Dean Brown in J Med Chem and follow-up posts by Keith Hornberger and Derek Lowe prompted me to think about how we train computational chemists and cheminformaticians for careers in drug discovery. It also brought to mind some unique differences between how computational chemistry is practiced in biotech and pharma. For those who haven’t read Dean Brown’s editorial and the subsequent reactions, I’d highly recommend them. In short, the authors focused on how medicinal chemists were trained in the past and how biotech and the growth of outsourcing are changing that model. Traditionally, most medicinal chemists received academic training in organic synthesis labs and then learned medicinal chemistry on the job from more experienced colleagues. Chemists would typically start at the bench and gradually transition to roles where they led groups and/or drug discovery project teams. With the rise of smaller biotechs and the advent of chemistry outsourcing, many medicinal chemists no longer have the same opportunities to learn through this apprenticeship model.

Learning the Ropes
Like medicinal chemists, most computational chemists don’t come into pharma or biotech with experience in drug design.  Computational chemists come from a range of academic backgrounds, including quantum chemistry, biophysics, molecular dynamics, docking, QSAR modeling, and machine learning.  Although some students work collaboratively with experimentalists on molecular design, these efforts typically only yield a handful of compounds. It is rare for graduate students to be exposed to tasks like optimizing pharmacokinetics or off-target binding.  Like medicinal chemists, most computational chemists learn on the job from more experienced colleagues. When someone joins a drug discovery company from academia, they are often paired with more experienced colleagues.  These mentors can help new employees learn the necessary scientific skills as well as the nuances of being part of a multi-disciplinary team.  As Dean Brown pointed out, these learning opportunities may be less available in a small biotech. Of course, large organizations can also have downsides.  In big pharma, it’s easy for a computational scientist to be tagged as a niche specialist and considered “the MD person” or “the machine learning person.” Subsequently, that can become all you do.  As I’ll discuss below, that’s not necessarily a bad thing.  Goals and definitions of success vary from person to person.  When joining a small biotech directly from academia, one may only have one or two colleagues to learn from.  However, in smaller companies, often out of necessity, one can have opportunities to explore a wide range of areas and roles, both within and outside of computation. The era of big pharma as a “job for life” seems to be coming to an end.  It has become common for people to start their careers in large organizations where they can be surrounded by mentors and then move into smaller biotechs that may provide different opportunities. 

To provide context, I started my industry career at a biotech with 140 employees.  Over the next 20 years, I saw that company grow to more than 2,000 people at three sites.  After that, I joined a 10-person startup, which has grown to more than 300 employees.  While I haven’t had the true big pharma experience, I think I’ve learned a bit about it from years of conversations with friends and colleagues.  I’ve been fortunate to work in companies with a strong emphasis on computation.  While the first company I joined had 140 employees, there were six other computational chemists, all of whom had significant industry experience. Having that group of mentors was critical for my professional development.  

Avoiding or Embracing Silos
As I mentioned above, computational chemists in large organizations can be viewed as specialists in a particular area and become siloed.  For some people, this can be a good thing.  If someone is passionate about a particular area and wants to become a domain expert, this can be valuable for them and the company.  There are a lot of rewarding aspects to developing new methods within a drug discovery company.  You get a firsthand opportunity to see your methods in action and can work collaboratively with experimentalists to improve those methods. However, as a specialist, it’s important to remember that advancement is based on impact. If you want to be promoted, you must ensure your work helps to advance drug candidates.  Drug discovery is about decisions, and your work, whether it’s on infrastructure, methodology, or direct project support, should inform decisions.

Not everyone wants to be a specialist.  Some computational chemists may aspire to eventually lead drug discovery projects or teams.  In these cases, broadening your skillset into other areas like medicinal chemistry, pharmacology, biophysics, pharmacokinetics, and disease biology is important.  In my experience, the most successful drug discovery project leaders are curious people with broad scientific knowledge.  There are many ways to broaden your skills. 
  • Read papers from areas outside your discipline. Mark Murcko’s 2018 paper in J Med. Chem provides a lot of great advice for drug hunters.  
  • Find mentors from other disciplines who can help you learn.
  • Attend courses; the Drew University ResMed course is a great place to start.  The Medicinal Chemistry Gordon Research Seminar (GRS) provides a condensed version of ResMed.  I’ve also heard good things about the Leysin, and RSC MedChem courses. 
  • Watch YouTube videos (a personal favorite). 
  • Ask a lot of questions. 
Building a Support Network 
In Dean Brown’s editorial, he talks about how pre-competitive collaboration can help biotech chemists learn to perform tasks that other groups in larger organizations might handle. Precompetitive collaborations are equally crucial for computational chemists in both biotech and pharma.  It’s incredibly useful to share information on which commercial and open source methods work best and discuss interesting papers.  There are numerous local interest groups, like the Boston Area Group for Informatics and Modeling (BAGIM), which has spawned similar meetups in New York (NYAGIM) and Southern California (SAGIM).   The San Francisco Bay Area has Comp Together, and I’m certain there are equivalents in other geographies.  These gatherings provide an opportunity to meet with local folks, share information, and hear great talks.  Conferences also provide an excellent means of establishing collaborations and making contacts.  My favorites include the Computer-Aided Drug Design Gordon Conference (CADD GRC), the OpenEye Scientific Software user group meeting (CUP), the Chemical Computing Group user group meeting (CCG-UGM), and the RDKit UGM.  Social media provides another great way to connect with your peers.  There are scores of excellent Comp Chem and Cheminformatics conversations on LinkedIn.  Social media can also let you connect and chat with people you might not otherwise meet. Sadly, there was a good community of ChemTweeps on Twitter, but Elon spoiled the party.  At this point, it’s unclear where that flock will migrate. 

Tools of the Trade
One aspect that some don’t consider when joining a small biotech company is the availability of commercial molecular modeling software.  Companies operating on limited budgets may be reluctant to spend money on expensive commercial software.  The leadership in these organizations may believe it’s possible to operate solely on freely available software.  While performing some tasks with open source tools may be possible, others will be challenging. For instance, if you’re building machine learning models using data from drug discovery programs, open source software will probably be as useful as anything you can purchase.  True, it typically takes a bit of programming skill to munge your data into the appropriate format and build a model, but it’s not very hard, and there are numerous examples that you can use as starting points. On the other hand, tasks like large-scale virtual screening are much easier when using commercial software tools. Docking millions of molecules with only open source software will quickly become a frustrating experience. It should be noted that not all biotechs skimp on software licenses.  Some have abundant software resources.  Before joining any company, it’s important to know before you go and ask about software resources and budgets.   You’ll be glad you did. 

Great Expectations
Another unique aspect of computation in biotech, and sometimes in larger pharma, is managing expectations.  The recent hype and widespread lack of understanding of AI can sometimes lead to unrealistic expectations from senior management.  On more than one occasion, I’ve had friends who work in biotech tell me, “Our board of directors is telling us we need to do more with AI.”  My reply is typically, “What problem are you trying to solve?” to which they answer, “I don’t know, but we need to do more with AI.”  Many organizations seem to look at AI as a train they need to get on instead of evaluating the critical problems and whether AI is an appropriate solution.  Of course, computational misconceptions are not exclusive to biotech; silly things happen in pharma, too.  On two separate occasions, VP-level medicinal chemists from large pharmaceutical companies told me, “We have a solubility model that is so good, we stopped running the assay.”  I’ve written about solubility prediction in the past.  It’s technically challenging from both an experimental and computational standpoint.  The probability that anyone has built the ideal model is very low. Unfortunately, in some cases, these “successes” are due to a lack of statistical rigor rather than scientific breakthroughs.  I think one of the most important skills for those in computational disciplines is “managing up.”   We must ensure senior leaders understand what is and is not possible.  For example, we must explain that large language models simply predict the next token in a sequence, and SkyNet is not waiting in the wings. In addition, contrary to Senator Chris Murphy's statement earlier this year, ChatGPT doesn’t understand Chemistry. 

In Conclusion
Given the smaller size of computational chemistry groups, the differences between biotech and pharma may be even larger for computational chemists than for medicinal chemists.  In a biotech with a couple of computational chemists, it may be challenging for new hires to learn the scientific and social skills necessary to be effective on the job. However, while larger organizations can sometimes provide a broader group of colleagues and mentors, it’s important to ensure alignment between your aspirations and those of the company.  Regardless of the organization's size, it’s helpful to understand the degree to which you can interact with the broader community. As we’ve moved into a world where computational work can be performed remotely, it’s important to consider the importance of broader scientific interactions in your professional development.  Identifying opportunities for mentoring a cross-functional interaction can be more difficult when you only see people on Zoom.  Finally, while drug discovery has a rich literature, many of the most important lessons are handed down through an oral tradition.  As our industry evolves, we must find ways to continue this tradition and enable the next generation of drug hunters. 

References
On the Medicinal Chemistry Pedagogy

Adapting to the Changing Landscape of Biotech-Driven Drug Discovery

The Current Industry Landscape

What Makes a Great Medicinal Chemist? A Personal Perspective



Comments

Post a Comment

Popular posts from this blog

We Need Better Benchmarks for Machine Learning in Drug Discovery

Image
Most papers describing new methods for machine learning (ML) in drug discovery report some sort of benchmark comparing their algorithm and/or molecular representation with the current state of the art.  In the past, I’ve written extensively about statistics and how methods should be compared .  In this post, I’d like to focus instead on the datasets we use to benchmark and compare methods.  Many papers I’ve read recently use the MoleculeNet dataset, released by the Pande group at Stanford in 2017, as the “standard” benchmark.   This is a mistake.  In this post, I’d like to use the MoleculeNet dataset to point out flaws in several widely used benchmarks.  Beyond this, I’d like to propose some alternate strategies that could be used to improve benchmarking efforts and help the field to move forward.   To begin, let’s examine the MoleculeNet benchmark, which to date, has been cited more than 1,800 times.   The MoleculeNet collection consists of 16 datasets divided into 4 categories.  Qua
Read more

AI in Drug Discovery 2023 - A Highly Opinionated Literature Review (Part I)

Image
Here’s the first part of my review of some interesting machine learning (ML) papers I read in 2023.  As with the previous editions , this shouldn’t be considered a comprehensive review.  The papers covered here reflect my research interests and biases, and I’ve certainly overlooked areas that others consider vital.  This post is pretty long, so I've split it into three parts, with parts II and III to be posted in the next couple of weeks.    I. Docking, protein structure prediction, and benchmarking II. Large Language Models, active learning, federated learning, generative models, and explainable AI III. Review articles 2023 was a bit of a mixed bag for AI in drug discovery.  Several groups reported that the deep learning methods for protein-ligand docking weren’t quite what they were initially cracked up to be.  AlphaFold2 became pervasive, and people started to investigate, with mixed success, the utility of predicted protein structures.  There were reports of significant advanc
Read more

Getting Real with Molecular Property Prediction

Image
Introduction If you believe everything you read in the popular press, this AI business is easy. Just ask ChatGPT, and the perfect solution magically appears. Unfortunately, that's not the reality. In this post, I'll walk through a predictive modeling example and demonstrate that there are still a lot of subtleties to consider. In addition, I want to show that data is critical to building good machine learning (ML) models. If you don't have the appropriate data, a simple empirical approach may be better than an ML model.  A recent paper from Cheng Fang and coworkers at Biogen presents prospective evaluations of machine learning models on several ADME endpoints.  As part of this evaluation, the Biogen team released a large dataset of measured in vitro assay values for several thousand commercially available compounds.  One component of this dataset is 2,173 solubility values measured at pH 6.8 using chemiluminescent nitrogen detection (CLND), a technique currently consider
Read more