Much has been written about the merits and demerits of the virtual biotech model. In essence, a virtual biotech outsources all non-core activities and rejects the need for in-house laboratories. Instead, a small internal team manages execution outside the company, generally through contract research organizations (CROs). Such companies have led to successful exits and IPOs – Ferrokin, Artaeus, Stromedix, NovaCardia, Stemline – but have principally focused around specific, clinical-stage assets.
To our knowledge, Nimbus Discovery is the first company to have established a drug discovery platform entirely through a virtual, distributed model. Nimbus conducts drug design and computational chemistry itself but has no in-house laboratories. Over the past four years, we have established a complete small-molecule discovery and development platform that spans early discovery through to early development. We have six on-going programs across three therapeutic areas, all managed by a lean team of twelve internal employees that directs approximately eighty external FTEs across three continents.
The rise of the virtual company
Virtual drug discovery was initially a response to the high cost of capital for startups and the significant burden of fixed infrastructure, but true enablement of the approach has only been possible comparatively recently: a rich CRO ecosystem is now in place giving small companies access to every conceivable R&D capability; massive Pharma layoffs have enhanced the accessible talent pool; CROs in lower-cost geographies such as China and India offer higher quality services with an improved user interface; and communication tools, such as inexpensive videoconferencing, have made cross-border collaboration much more feasible.
Moreover, the successful adoption of the distributed model in other industries provides a blueprint for the life sciences. In the semiconductor industry, ARM Technologies designs microchip layouts that use markedly fewer transistors than traditional approaches offering speed, efficiency and cost advantages. Unlike its fully-integrated competitors, ARM does not manufacture its own chips. Instead, it licences its designs to hundreds of chip manufactures worldwide, receiving high-margin royalties in exchange. ARM-designed chips are now present in 95% of smart phones, highlighting how pervasive a company without a manufacturing base can become.
Successes and surprises
Nimbus Discovery was designed to push the virtual approach to the limit – a “radically virtual operating model”, as we called it – based on the assumption that it would be better, cheaper and faster than bricks-and-mortar approaches. At the time, naysayers argued that the approach would be disadvantaged from a speed and data quality perspective due to perceived loss of control and coordination.
But has the model worked out as everyone expected? Well, there have been some successes and some surprises.
Nimbus has focused on largely intractable targets where the rest of the industry has struggled to achieve clinically-viable drug candidates due to poor potency or selectivity, off-target safety risks, or inadequate drug-like-ness. Despite this, Nimbus has brought two programs to Development Candidate (DC) stage for targets where many others have broken their pick axes. This, I believe, provides substantial validation for the Nimbus model and our underlying technology.
In benchmarking against other portfolio companies, I feel that we hit the ball out of the park for our program targeting the lipid master regulator, Acetyl CoA Carboxylase (ACC), where we were able to go from virtual screen to DC in 16 months. Our IRAK4 program delivered a Lead Candidate four months after virtual screen but we then spent substantially longer improving pharmaceutical properties.
Our model appears to be far cheaper and faster early in discovery, where generating superb drug-like hits and leads is the goal, but in the latter stages of drug discovery, such as final lead optimization and candidate selection, the costs and time we have faced look a lot like other quality biotech campaigns around ADME, toxicology, PK, etc. If you have a starting point without downstream liabilities in those areas, often proven empirically, it translates into some startlingly good metrics. This teaches us how we can further improve our time/cost to DC metrics: by ensuring that we have multiple, structurally distinct scaffolds to take into Lead Optimization, we can choose the best to expedite our path to DC.
As previously noted, we have a different cost structure to traditional biotech. Only 10-15% of our cost is fixed and most of the variable cost is spent with external CRO’s that can titrated up and down at very short notice. This allows us to allocate resource dynamically based on program requirements on a month-to-month basis. This is in contrast to high fixed-cost models where decision-making is driven by how to deploy, and pay for, the annual costs of buildings, labs and an installed employee base. Feeding the infrastructure becomes an important part of those models. Since the vast majority of our resources at Nimbus are fungible, we can make business decisions that are truly data driven.
Looking under the hood
So how has this been achieved and where do we still have hurdles to overcome in operating this model? I believe that the key success drivers for the Nimbus model are people, programs and partners.
The Nimbus drug discovery platform consists of a distributed ‘network’ of team members that can efficiently share ideas, information, data and compounds. This has allowed us to cherry pick the best talent across the globe rather than relying solely upon the talent pool in Cambridge, Massachusetts. Similarly, we have recruited the world experts around certain target biologies to run key aspects of programs, in a way that they feel truly part of the team. As a Brit, it gives me great warmth to say, “the sun never sets” on the Nimbus team.
Program leaders at Nimbus are then tasked with marshaling resources outside the company to generate the required datasets and meet aggressive timelines. This requires well-seasoned R&D executives that can design, and then manage, science at a distance. Scientists that thrive within the Nimbus model are strong individual contributors that have a collaborative spirit and can motivate through influence rather than direct reporting relationships.
By outsourcing non-core activities, we reduce the footprint of the organization allowing our scientists to focus on doing science rather than addressing HR issues and writing performance reviews. There’s also no room in the Nimbus model to build an empire … but this is precisely the point. Empires lead to silos and politics, both of which compromise good decision-making.
This type of approach is not for everyone. Most scientists have been trained to trouble shoot experiments that they can touch and feel. The distributed model places the control of execution in the hands of others, often many time zones away and it takes a significant time commitment to manage these external relationships. Also, many senior executives are used to the infrastructure associated with larger companies, none of which is present at Nimbus. The Nimbus model demands that the entire team roll up their sleeves and regularly ‘helicopter’ from high-level strategy to the details of experimental design.
Since we work so differently at Nimbus, finding the right people has been challenging and, given the small size of the organization, every hire must be vetted extremely carefully. That said, we have been able to attract a cadre of seasoned R&D veterans that think differently, an important differentiator given Nimbus’ focus on previously intractable targets.
Despite the success of the Nimbus model, I do not believe that it is appropriate in all cases. Nimbus principally works with well-understood but historically challenging targets, known modalities and established chemistries. Assays, experiments, and chemistry can all be accessed creatively in the external marketplace but I would not try to implement a novel RNAi, peptide or biologics platform using the Nimbus model. For these approaches, the core value proposition is based on proving out a new therapeutic modality to create entirely novel biological underpinnings. Such skills are, by definition, not available within CROs.
In situations where we have had to explore new biology, e.g. IRAK4’s role in tumorigenesis, we have adopted a more standard model that involves some rented lab space and some talented biologists. At times, we have also enlisted the help of US-based chemists to solve hard synthetic chemistry problems as we scaled synthesis for IND-enabling studies.
The commitment of strong technology partners is key to the success of the Nimbus model. Schrodinger, a key technology and co-founding partner, has been instrumental in the success of Nimbus and are core team members in the “engine room” of the company. Beyond computational insight, we’ve embraced dozens of other partners. This has involved careful screening of potential collaborators but also a healthy amount of empiricism.
Frankly, we still find it challenging to find true full-service CROs, and our bias continues to be towards cherry picking capabilities from the best global CROs but this obviously comes with a complexity burden. To address these issues, we have looked for creative ways to align incentives and share risk. To make sure that we get the A team working on our projects, we have implemented success-based milestone arrangements; preferred-partnership agreements that ensure enhanced access to services and turn around based on volume of work; and bonus awards to individual scientists.
Despite tremendous growth in the ecosystem, there are still areas where the CRO footprint is more limited. On the biology side, Asian CROs are adept at primary assays and off-the-shelf ADME readouts but often struggle with bespoke biology. This is compounded by import restrictions into China for assay kits and lack of infrastructure for donor tissues, both of which are commonplace in the US. This situation will undoubtedly change over time and we are aware of certain Asian CROs that are now establishing US-based teams to develop custom assays which can then be transferred to China once protocols are robustly established.
For complex biology, we have also made extensive use of academic collaborations. This should be a blog post in its own right but we have found that these relationships can generate pivotal datasets and, if successful, can build excitement amongst the medical community.
Beyond capital efficiency
The Nimbus operating model was initially conceived as a route to improving capital efficiency and productivity for tackling hard-to-drug targets, which it has clearly delivered on … but the reality of the model, I believe, goes well beyond that and is much more exciting. We have established a new way of doing drug discovery that redefines how we innovate. The Nimbus approach is inherently more nimble and flexible, and unencumbers scientists to think about science rather than organizational issues. Nimbus is as much about the mindset of its team as it is about the underlying technology. This mindset is expressed through the configuration of our organization: function follows form in the same way that flexible thinking maps to fungible resources.
For the right type of projects, I believe that the “Nimbus experiment” in virtual R&D represents an important drug discovery model for the future.