ChatGPT is taking over. Or something like that.

The reality is more nuanced. As the old saying goes, "shit in, shit out." That summarizes the mathematical probabilities behind large language models (LLMs) like ChatGPT. It could also foreshadow bottlenecks and speed limits for popular tools, which may become evident just as quickly as the mostly-deserved hype for the tools themselves.

The bottlenecks going largely undiscussed right now hinge on a single nuance: Large datasets aren't valuable. Large datasets that are properly annotated are valuable. That's likely to wreak havoc and make fortunes for biotech investors this decade.

Although that doesn't bode well for every company in our coverage ecosystem, it strengthens the thesis for Relay Therapeutics. It doesn't really have anything to do with ChatGPT directly, but the LLM's impressive performance in certain tasks made us revisit our model for the business.

POS In, POS Out

The single-most important metric for precommercial drug developers is the probability of success (POS).

This is typically defined as the percent chance a clinical-stage asset eventually earns regulatory approval. This metric weighs heavily on the numbers churned out by a risk-adjusted net present value (rNPV) model, such as those we build at Solt DB Invest. There are nuances, though.

The POS changes depending on the stage of development. According to a report from QLS Advisors and BIO summarizing industry averages from 2011 to 2020:

• An asset in phase 1 development has a 7.9% chance of earning approval.
• An asset in phase 2 development has a 15.1% chance of earning approval.
• An asset in phase 3 development has a 52.4% chance of earning approval.
• An asset submitted to the U.S. Food and Drug Administration (FDA) following successful clinical development has a 90.6% chance of earning approval.

The POS also changes depending on the therapeutic area. According to the same summary report:

• A urology asset in phase 1 development has a 3.6% chance of earning approval.
• An oncology (cancer) asset in phase 1 development has a 5.3% chance of earning approval.
• A hematology (blood) asset in phase 1 development has a 23.9% chance of earning approval.

The POS metric isn't only used to estimate the odds of earning regulatory approval. There are phase-to-phase transition rates, too.

• An oncology asset in phase 1 development has a 48.8% chance of advancing to phase 2 development.
• An oncology asset in phase 2 development has a 24.6% chance of advancing to phase 3 development.
• An oncology asset in phase 3 development has a 47.7% chance of advancing to regulatory submission.
• An oncology asset submitted to regulators has a 92% chance of earning approval.
• If you multiply all these percentages together, then you get the phase 1 to approval POS of 5.3% for oncology assets (0.488 x 0.246 x 0.477 x 0.92 = 0.053).

As implied above, investors can get very granular. For example, the POS also changes depending on the therapeutic modality – Fc engineered antibodies, AAV gene therapy, GalNAc-siRNA, etc. – but that's part of our secret sauce at Solt DB Invest.

Let's bring this back to Relay Therapeutics. The precommercial drug developer is initially focusing on oncology assets. However, an asset it moves into phase 1 clinical development will have a significantly higher approval POS than 5.3%.

The company's Dynamo technology platform combines experimental data and computational data to simulate protein motion. Meaning, Relay Therapeutics takes accurate 3D protein structure models (perhaps from AlphaFold2) and simulates how they wiggle and move across time. These 4D protein structures, with time representing the fourth dimension, unveil intricate details about binding pockets (where a drug compound attaches to disable or correct a disease-driving protein).

The insights could set a new standard for the POS of oncology assets.

Forecast & Modeling Insights

For example, let's assume the high-quality, proprietary datasets from Dynamo lead to more selective drug compounds with significantly fewer side effects than existing treatment options for the same genetic target. Considering many drug candidates in phase 2 fail due to safety issues, that would lead to a significantly higher transition rate for the company's mid-stage assets.

• If Relay Therapeutics advances half of all oncology assets from phase 2 to phase 3 clinical trials, but realizes the industry average transition rates for all other stages of development. That alone would lift the pipeline's phase 1 to approval POS to 10.7% (0.488 x 0.500 x 0.477 x 0.920 = 0.107). That's double the industry average.
• If Relay Therapeutics also advances two-thirds of all oncology assets from phase 3 to regulatory submissions, then it would realize a phase 1 to approval POS of 15.0%. That's nearly triple the industry average.

Here's the thing: Highly-selective drug compounds are more likely to skip phase 3 development altogether. If the FDA grants Breakthrough Therapy designation, or allows a phase 2 study to serve as the pivotal trial required to earn regulatory approval, then the last two transition rates would count as 100% in the calculation. Combining the beneficial impact of cleaner safety profiles and smoother regulatory pathways could lead to a phase 1 to approval POS of well over 25%, or nearly five times the industry average.

That's not all. Relay Therapeutics designs ridiculously intricate phase 1 studies with dozens of meticulously-screened patients, multiple patient populations, and finely-tuned dosing levels. Investors should expect ruthless early-stage go / no-go decisions as a result. In other words, the company won't hesitate to terminate development of phase 1 assets that fail to meet a high internal data standard, but assets that do make it to phase 2 development would likely have an approval POS approaching or exceeding 50%.

That's not unheard of in drug development. RNA interference (RNAi) drug candidates targeting the liver with GalNAc ligands have a phase 1 to approval POS of nearly 60%, which is by far the highest of any therapeutic modality. But even a double-digit approval POS for early-stage assets would represent a paradigm shift in oncology.

Two near-term de-risking events could bolster our updated model:

• On Friday, April 14, preliminary phase 1 data for RLY-2608 (PI3K-alpha) will be published when an abstract for the American Association for Cancer Research (AACR) Annual Meeting 2023 becomes public. Relay Therapeutics is developing multiple drug candidates for PI3K-alpha, which could become the most valuable collection of assets in the global drug development pipeline. It appears the company may terminate RLY-2608 and focus on RLY-5836, a chemically distinct compound targeting the same mutations, but the data unveiled in mid-April could still be best in class. If the data are impressive, and the fast follower asset is assumed to be even more selective, then the company's valuation would need to reflect that.
• A complete phase 1 data readout for RLY-4008 (FGFR2) is expected in the second half of 2023. Preliminary data revealed the lead drug candidate demonstrated an overall response rate of 88% at the pivotal dose in a specific patient population, which is more than double the response rate of the leading drug product on the market. The FDA has allowed a phase 2 clinical trial to serve as the pivotal study, which is slated to complete enrollment before the end of the calendar year. However, investors can assign a high likelihood that the asset earns FDA Breakthrough Therapy designation and shouldn't be surprised if it earns accelerated approval by the end of 2024.

Margin of Safety & Allocation

(Increased)

Relay Therapeutics is considered a Growth (Quality) position. The current Margin of Safety range for the company based on our 2023 model is below:

• Current Price (market close April 4):  $16.89 per share
• Likely Undervalued:          <$22.47 per share
• Midpoint:                            $26.22 per share
• Likely Overvalued:            >$29.96 per share
• Allocation Range:              Up to 15%

Relay Therapeutics reported 121.384 million shares outstanding as of February 17, 2023. The Margin of Safety range above assumes 133.523 million shares outstanding to account for 10% dilution expected from secondary offerings in the next 24 months.

The previous Margin of Safety range spanned $16.29 per share to $24.43 per share with a midpoint of $20.36 per share.

Further Reading

• February 2023 press release for fourth-quarter and full-year 2022 highlights
• January 2023 press release from Eli Lilly announcing FDA approval of Jaypirca based on phase 1/2 data alone. The asset has many parallels to the development of RLY-4008.
• September 2022 research note reviewing additional phase 1 data for RLY-4008
• September 2022 research note reviewing the preliminary phase 1 data readout for RLY-4008

‍