future of pharmaceutical industry

What will the future of pharmaceutical industry look like?

It is clear that we are reaching one end of a paradigm, but what most people still do not get is how big the oncoming changes will be. We are on the cusp of a great intellectual revolution, on par with the revolution in 20th century physics. Computer science is unlocking biology, just like mathematics unlocked physics, and the consequences will be huge. (Read this older post for a deeper look at this interesting analogy between analogies.)

For the first time in history, we are engineering solutions from scratch rather than stumbling into them or stealing them from nature. Western medicine is only now truly taking off.

Not only will this transformation be breathtaking, but it will also be unfolding at a speed much faster than we expect. As biology becomes more information theoretical, pharmaceutical industry will become more software driven and will start displaying more of the typical dynamics of the software industry, like faster scaling and deeper centralization and modularization.

Of course, predicting the magnitude of change is not the same thing as predicting how things will actually unfold. (Sometimes I wonder which one is harder. Remember Paul Saffo: “We tend to mistake a clear view of the future for a short distance.”) Let us give a try anyway.


1. Splitting and Centralization of the Quantitative Brain

Just like the risk analytics layer is slowly being peeled out of big insurance companies as it is becoming more quantitative (small companies could not harbor such analytics departments anyway), the quantitative layer of the drug development process will split out of the massive pharmaceutical companies. (Similarly, in the autonomous driving space, companies like Waymo are licensing out self-driving technologies to big car manufacturers.)

Two main drivers of this movement:

• Soft Reason. Culturally speaking, traditional (both manufacturing and service) companies can not nurture software development within themselves. Big ones often think that they can, but without exception they all end up wasting massive resources to realize that it is not a matter of resources. Similarly, they always end up suffocating the technology companies they acquire.

• Hard Reason. Unlike services and manufacturing, software scales perfectly. In other words, the cost of reproduction of software is close to nil. This leads to centralization and winner-takes-all effects. (Even within big pharmas bioinformatics and IT departments are centralized.) Software developed in-house can never compete with software developed outside, which serves many customers, takes as input more diverse use cases and improves faster.

Study of complex systems (which biology is an example of) is conducted from either a state centric or process centric perspective, using either statistical (AI driven) or deterministic (algorithm driven) methods. (Read this older post for a deeper look at the divide between state and process centric perspectives.)

In other words, the quantitative brain in biology will be centralized around four different themes:

1. Algorithm Driven + State Centric

2. AI Driven + State Centric

3. Algorithm Driven + Process Centric

4. AI Driven + Process Centric

Xtalpi is a good example for the 4th category. Seven Bridges in its current form belongs to the 1st category. There are other examples out there that fit neatly into one of these categories or cut across a few. (It is tough to cut across both state centric and process centric perspectives since latter is mostly chemistry and physics driven and tap into a very different talent pool.)


2. Democratization and Commodification of Computation

Big pharma companies could afford to buy their own HPCs to run complex computations and manage data. Most are still holding onto these powerful clusters, but they are all realizing that this is not sustainable for two main reasons:

• They either can not accommodate bursty computations or can not keep the machines busy all time. So it is best for the machines to be aggregated in shared spaces where they are maintained centrally.

• Since data size is exploding doubly exponentially, it is becoming harder to move and more expensive to store. (Compute needs to go where data is generated.)

Cloud computing took off for reasons entirely unrelated to biomedical data analysis, which will soon be the biggest beneficiary of this revolution as biomedical data sizes and computation needs surpass everything else. (It is not surprising that the centralized disembodied brain is developing in the same way as our decentralized embodied brains did. It got enlarged for social reasons and deployed later for scientific purposes.) Small biotechs can now run complex computations on massive data repositories and pay for computation just like they pay for electricity, only for the amounts they use. Big pharmas too are migrating to the cloud, finally coming to terms with the fact that cloud is both safer and cheaper. They are no longer uncomfortable departing with their critical data and no longer ignorant about the hidden costs of maintaining local hardware.

Long story short, democratization of computation is complete (aside from some big players with sunk cost investments) and the industry has already moved on to its next phase. Today we are witnessing a large scale commoditization of cloud services, driven by the following two factors:

• Supply Side. Strong rivals arriving and catching up with Amazon Web Services.

• Demand Side. Big players preferring to be cloud agnostic and supporting multi-cloud.

3. Democratization, Uniformization and Centralization of Data

Democratization. Big pharmas are hoarding data. They are entering into pre-competitive consortiums and forming partnerships with or buying diagnostics companies straight out. Little pharmas (startup biotechs) are left out of this game, just as they were left out of the HPC game. But just like Amazon democratized computing, National Institutes of Health (NIH) is now trying to democratize data. (Amazon and NIH are playing parallel roles in this grand story. Interesting.) Sooner or later public data will outstrip private data simply because health is way too important from a societal point of view.

Uniformization. NIH is also trying to uniformize data structures and harmonize compliance and security standards across the board, so that data can flow around at a higher speed.

Centralization. NIH not only wants to democratize and uniformize data, but it also wants to break data silos. Data is a lot more useful when it all comes together. (Fragmentation problem is especially acute in US.) Similarly, imagine if everyone could hold all of their health data on a blockchain that they can share with any pharma in return for a compensation. This is another form of centralization, radically bringing together everything at an individual level. All pharma companies need to do is to take a cross section across the cohorts they are interested in.

With its top-down centralized policy making and absence of incumbent (novel drug developing) big pharmas, China will skip all of the above steps just as Africa skipped grid-based centralized electricity distribution and is jumping straight into off-grid decentralized solar power technologies.


4. Streamlining and Cheapening of Clinical Trials

It is extremely time consuming and expensive to get a drug approved. In 2000s, only 11 percent of drugs entering phase 1 clinical trials ended up being approved by FDA. Biotech startups that can make it to phase 3 usually end up selling themselves completely (or partially on a milestone basis) to big pharma companies simply because they can not afford the process. In other words, the final bottleneck for these startups in getting to the market on their own is clinical trials.

This problem is much more multi dimensional and thorny, but there is still hope:

• Time. Regulations are being more streamlined and thereby making the processes faster.

• Cost. Genomics and real world data are enabling better targeting (or - in the case of already approved drugs - retargeting) of patients and resulting in better responding cohorts and thereby driving costs down.

• Risk. As we get better at simulating human biology on hardware and software, parallelizability of experimentation will increase and thereby the number of unnecessary (sure to fail) experiments on human beings will decrease. In other words, just as in the software world, experiments will fail faster.

5. Democratization and Decentralization of Drug Development

As some of the largest companies in the world, big pharmas are intimidating, but from an evolutionary point of view, they are actually quite primitive. The existing fatness is not due to some incredible prowess or sustained success, it is entirely structural in the sense that the industry itself has not fully matured and modularized yet. (In fact, there is little hope that they can execute the necessary internal changes and evolve a contemporary data-driven approach to drug development. That is why they seek acquisitions, outside partnerships etc.)

If you split open a big pharma today, you will see a centralized quantitative brain (consisting of bioinformatics and IT departments) and a constellation of independent R&D centers around this brain. This is exactly what the whole pharma industry will look like in the future.

Once quantitative brain is split off and centralized, computation is democratized and commoditized, data is democratized, uniformized and centralized, and clinical trials is streamlined and cheaper, there will be no need for biotech startups to merge themselves into the resource-rich environments of big pharma companies. Drugs will be developed in collaboration with the brain and be co-owned. (Currently we have already started seeing partnerships between the brain and the big pharma. Such partnerships will democratize and become common place.)

Biology will start off in independent labs and stay independent, and the startups will not have to sell themselves to the big guys if they do not want to, just as in the software world.

Biology is way too complex to allow repeat successes. Best ideas will always come from outsiders. In this sense, pharma industry will look more like the B2C software world rather than the B2B software world. Stochastic and experimental.

We have already started to see more dispersed value creation in the industry:

“Until well into the 1990s, a single drug company, Merck, was more valuable than all biotech companies combined. It probably seemed as if biotech would never arrive—until it did. Of the 10 best-selling drugs in the US during 2017, seven (including the top seller, the arthritis drug Humira) are biotech drugs based on antibodies.”

- MIT Tech Review - Look How Far Precision Medicine Has Come

(I did not say anything about the manufacturing and distribution steps since the vast majority of these physical processes is already being outsourced by pharma companies. In other words, these aspects of the industry have already been modularized.)