Deciphering the Rules for Engineering Precision Medicines

At Creyon, we aspire to engineer novel precision medicines (Oligonucleotide-Based Medicines or OBMs) with predictable performance in the clinic, completely eschewing trial-and-error screening. OBMs should be engineerable because they are polymers with compositional units and interactions between these units—very similar to this sentence. Unlike this sentence, however, oligonucleotides are expensive to synthesize, and the structure of these “words” is not obvious. Moreover, clinically relevant pharmacology datasets that can confidently report on the probability of human safety and efficacy are very expensive to create. That makes it a challenge to unravel the molecular rules to engineer OBMs.

So how do we make such engineering possible? We must completely rethink how to gather pharmacology datasets in order to decipher the hidden code of what makes a safe and efficacious oligonucleotide. Big data is impossible to gather in this context—the design space of these molecules is astronomically large. An oligonucleotide monomer is made of three Lego blocks—linker, sugar and base. Each of these blocks can be modified by tens to hundreds of possible nucleic acid modifications reported in the literature over the last several decades. These modifications were typically developed to improve pharmacology properties of otherwise easily-degradable native RNA or DNA-like molecules, for example, their biodistribution or tissue half-life.?

One of the biggest challenges in the field is: how can we rationally utilize these modifications to control the pharmacology properties of OBMs? The number of possible combinations of monomers is so large!? The field therefore focuses on a few modifications, or even single modifications at single positions of a set of oligonucleotides to make sense of the impact of these modifications on pharmacology. Often, reckless generalizations are made in the industry from such limited experimentations. For example, the literature on ASOs and siRNAs often claims universally better chemistry or architecture based on the flimsy support of a few ad hoc modifications on a few sequences.

But why? In 1971, Ronald Fisher in his treatise “Design of Experiments” bemoaned— “In expositions of the scientific use of experimentation it is frequent to find excessive stress laid on the importance of varying the essential conditions one at a time.”?

We need to gather data smartly. We need to design better experiments that ask questions efficiently. In the accompanying work, we present several algorithms in Design of Experiments (DoE). These designs can query how independent, pairwise and higher-order contributions of the monomer of OBMs contribute to pharmacology in a minimal set of experiments, not one at a time. This minimal set also avoids bias and systematic errors while obeying constraints that may arise from experimental conditions or chemical synthesis limitations.?

We have used these approximate algorithms (and a larger class of proprietary methods) at Creyon to repeatedly design optimal sets of oligonucleotides and test them in thousands of detailed animal pharmacology experiments. These detailed pharmacology experiments have led to unraveling the design rules for us—today we can design a hundred ASOs and eighty to ninety of them are typically safe in? liver, kidney and Central Nervous System (CNS) etc. in rodent pharmacology. The positive predictive value of our machine learning models built on such hyper-informative datasets is unprecedented in the field. In contrast, other companies working in the field boast of screening thousands of ASOs to arrive at a handful of suboptimal leads. The tools we present in the paper also have broader applicability in designing better copolymers and are not limited to oligonucleotides.?

Collectively, spanning our recent pioneering work on models and methods of off-target analysis, patents on our methods for machine learning in oligonucleotides, and the current work on Design of Experiments, we are systematically chipping away at the challenges of creating precision medicine on demand, tailored to the unique genetic profiles of patients all while demanding higher standards of safety and therapeutic index from the industry and ourselves. Stay tuned for more to come! We remain steadfast on the path to deliver oligonucleotide-based medicines more efficiently, transforming the paradigm from drug discovery to drug engineering.

Read our latest work—Efficient Approximate Methods for Design of Experiments for Copolymer Engineering.