Decentralization and Democratization of Drug Discovery:
“The future is not defined. It does not automatically happen. You have to make it happen.”
I’ve always found it really helpful to think about challenging problems in big ways. Thinking about what things look like in 10 years, then 5, helps me think about what to do next week. So today I’ll just share what’s been on my mind lately. I’ve been writing deeply about some of the topics I’ll mention, and probably have about 15 essays (chapters) to share in the future. But for now, I just want to talk about the ‘5/10 year’ views.
When I was 22, I had a dream about printing presses for biological data. I had no idea about integrated systems or high throughput screening, but this abstract concept arrived in my mind and never left. Over the past few years I’ve had a chance to physically work with these systems, understanding the opportunities and challenges of automated, scaled data collection. So far, we have organizations who see these processes like the mass production of artisanal crafts. Instead of filling a warehouse with an army of human basket weavers, we should build our own weaving machines to make baskets that no one else can.
For the age of TechBio, and for the ages to come, we need to move from individualized data generation, to team-oriented data manufacturing. What this requires are methods to standardize the preparation of reagents (ie. Media and cells etc) and proprietary tools specifically designed to capture phenotypes of interest, reproducibly. We are well aware of the reproducibility crisis in science, and while some of this may be due to bad note keeping, I can’t help but wonder whether this is also caused by the lack of standardization in lab plastics, reagents, and commercial measurement tools that aren’t suited for high throughput data collection. What if we poured our own multi well plates and built our own sequencers?
The printing press example was just a dream…. But here is my 5 year vision: farm-to-table drug discovery. Companies with proprietary phenotypes that are valuable to medicinal chemists, that are only accessible because of their custom built measurement tools. If the phenotype is actually valuable to medicinal chemists, then your pipeline will be gushing with meaningful assets in less time than it would take for your more artisanal (or semi-artisanal) peers to do so. Using the money you earn, you can acquire smaller reagents/ plastics players and eventually have total control over the data generation process. With this method, you have the chance to generate the best data (objectively)!
It seems obvious to me that companies that operate this way will remain in business because of the asset sales from internal pipelines, subscription fees from external chemists who access the platform, and other creative partnering. The other benefit specific to the platform, is that you can de risk the phenotype much faster if a lot of programs rely on it. A major (and unspoken) bottleneck in drug discovery is the number of medicinal chemists that can access a given platform. Every modern platform likely has 5-20 chemists surrounding it, each with their own opinions of the value of the data being generated. My prediction is that as these early companies with valuable phenotypes emerge, chemists from all over will swarm towards them by accessing the data through software. So, this vision of farm-to-table drug discovery is actually a vision of decentralized science.
To zoom in for a second, these companies have a name. They are called MechBios:
These are TechBio companies building at the interface of engineering and biology, upstream of data science and chemistry. They rely on mechanical, electrical, and chemical engineering to develop proprietary wet lab processes and instruments, to maximize signal and minimize batch effects in high throughput data collection. While TechBios may use commercial tech at scale for screening, MechBios use their custom tools for both screening and optimization. Anything that is observable in biology is a product of the biological response (model and chemical stimulus) and the tool to measure it. So, the goal would be to find a biological response (phenotype) that is valuable to chemists, and then build measurement technologies that are uniquely suited to capture that phenotype.
And to zoom out, for the real 10 year vision, what if we could relieve a major bottleneck of drug discovery by making anyone a chemist? We’ve seen the example of eteRNA, which was an RNA-folding game that regular people played to give biologists better RNA sequences to synthesize and test in their labs. What if these platforms with valuable phenotypes were not only buying reagents and plastics companies, but also paying small rewards to members of the public that had proposed useful therapeutic designs (according to some rules)? Could MechBios shepherd in a therapeutics revolution driven by harnessing not only elite engineers and data scientists, but the unique, diverse, collective intelligence of humanity?
I think this is going to happen. If the right steps are followed, MechBios can use public designers to move structures from plate to patient in less time than ever. Just think about an internal program with the support of hundreds of additional brains working through the night. Obviously these compounds need to be synthesized and tested, so the ‘rules’ described before may limit the complexity of molecular designs so that they can be made within a day for example. If we don’t need to disclose targets or disease indications, and instead just discuss some objective ‘optimization criteria’ that can be solved through a simplified representation of the therapeutic structure, then we will really be able to change the world, chemically.
I said I’d keep this essay to the 5/10 year plans, but there is actually a 15 year plan that I think about…
While flying home to Boston recently, I sat next to an active fund manager. As he described his work, I started to realize that it closely resembled the way medicinal chemists manage drug discovery optimization cycles. Just like active fund managers, chemists rely on a mix of quantitative analysis and some intuition/ previous experience to make decisions about what tests to run next. He described his fear of being automated out of a job by ‘passive’ methods, that directly relate tested investment inputs to fund performance outputs. Passive funds do well in conventional industries and well observed market conditions, while active managers have an advantage in unpredictable market conditions and niche industries.
Drug discovery data is collected in a vacuum, meaning that there aren’t unpredictable external factors to consider. And, niche markets could be analogous to novel (undiscovered) therapeutic modalities… where human inputs will initially be required. Humanity will benefit immensely from drug discovery’s transition to a passive (fully autonomous) industry, where MechBio platforms have learned to design, make, test, and analyze compounds towards beneficial ends without any human input. Eventually, the platform will hypothesize a table of structures with some known/ hypothesized effects that Pharma/ PE/ VC firms can buy, synthesize at scale, test in translatable models, and move into clinical trials. There will be many rewards for chemists and non-chemists to collect on the way to this autonomous future.
Drug discovery’s journey towards a fully autonomous future will be accelerated through the aggregation of tested inputs by 1) attracting chemists across the industry and 2) making anyone a chemist.
Also, thanks to Lee Cronin for the timely (tweeted at 4pm July 31st) introductory quote :)