Axial - Matrix #1 (de novo design)
Unique business models in life sciences
Axial partners with great inventors creating unique business models. Profiling exciting life sciences companies at the earliest stages is important. Rather than talk about their work specifically conveying the opportunity set is more important where the company and others in the field will bring to market currently confidential inventions to more people.
De novo design
Using AI and compute further down the central dogma is increasingly easier because of the super-exponential decrease in the cost of sequencing. There remain significant bottlenecks in RNA and protein analysis. So new companies can solve these new problems. Until the Illumina equivalent of mass spectrometry emerges (i.e. it could be InterVenn or Newomics) and the underlying circuitry of the cell becomes more predictable (i.e. hopefully Asimov with Cello can pull it off), there remain many niches to gain technical and business edges. One case of this is de novo design of proteins. The key advantage is mainly around specificity. Natural proteins have a tendency to interact with many targets. Depending on use that could be a feature or a bug. To solve this problem, one could search existing evolutionary designs or make new ones. Let’s talk about the latter. Although, evolution is probably the world’s greatest inventor.
De novo design doesn’t need to benefit from some sort of massive improvement in protein sequencing. Rather it just rides the increasing power of compute. Companies like Macromoltek, IgC, Neoleukin, and even DeepMind (probably the UK’s most important company right now) develop computational systems from a set of biophysical rules to design proteins from scratch. These proteins can be monoclonal antibodies, fragments, a natural protein (i.e. IL), or something else. Implicitly there are no limitations on what can be created - but it is a daunting data set to tackle (i.e. for only a 100 amino acid protein, there at 20^100 possible sequences) that could reveal new modalities that nature hasn’t discovered. This is actually a pretty controversial point - one side would argue that evolution could not have sampled every single protein, so something valuable has to be in the data and another side would take the position that most of the protein variants are functionally not important in cells. Axial would lean more toward the latter mainly due to prior work with David Haig, Guido Guidotti, Gary Ruvkun, and a few others whose work have shown the power of evolution. From a business perspective, de novo design still has a lot of value beyond unique designs - increased specificity and shorter design cycles.
Using this approach, currently intractable targets (i.e. undruggable genome) can be pursued and protein therapeutics can stand on their own as mono-therapies rather than relying on some sort of combination approach (i.e. there are over 1000 trials combining a molecule(s) with Keytruda or Opdivo - approved checkpoint inhibitors).
The strongest moats for businesses in the field probably will not be the initial data sets of proteins because they become increasingly public or the compute resources as they become more accessible. At this moment though, these can be used to bootstrap a unique model. Instead, the durable moats will be around the rules and the sets of experiments done. For de novo design businesses, the rules to search through the protein space coupled to the validation experiments will create platforms that become increasingly hard to catch. Companies using evolutionary methods probably win. For these computational fishing expeditions, knowing where to start is important like starting with an IL scaffold.
David Baker from University of Washington is the godfather, boss, consigliere all rolled into one of de novo protein design. Rosetta is one of the world’s most important inventions to accurately predict protein structure from primary sequence. David Baker likely talks to God every night where he receives another piece to add to the model to improve accuracy ever so much. He put out a great review a few years ago and the figure below comes from it and does a great job mapping three of the major methods for protein design: structure prediction, fixed-backbone, and de novo:
Reading the review is a more efficient way to understand the current technical progress and limitations. In short, structure prediction has improved massively, but balancing out stability with new functional sites is still a problem. The core issue is designing proteins with multiple low-energy states. Moreover, de novo methods need to do a better job incorporating hydrophobic pockets and enzyme active sites into their designs. This is where operating on natural proteins has many advantages. So long-term, technically using both evolution and de novo approaches probably wins. But there still is tension between the two camps making the individual incorporating these two paradigms very special.
Have a nice day.