#276: Cut It Out! Welcome to the beTTR Era of In Vivo Gene Editing, & More

At the Peripheral Nerve Society Conference this weekend, Intellia Therapeutics made history by presenting the first data of a CRISPR Cas9 based in vivo gene editing therapy in patients with Hereditary Transthyretin Amyloidosis (hATTR). Please find the results of this milestone trial in this New England Journal of Medicine publication.

hATTR is a disease in which amyloid proteins build up and, if untreated, can cause multiple organ failures. Typically, the primary impact is on the heart and nerves.

Until now, patients with hATTR faced limited curative treatment options. Alnylam’s Patisiran knocks down serum TTR by 80%, but greater knock down rates could enhance the clinical benefits.

Intellia Therapeutic’s presentation highlighted that editing the TTR gene could result in a one-time treatment, eliminating the need for chronic therapy. In a small six-person dose escalation study, the company showed that, in the second cohort at 0.3 mg/kg, three patients achieved an 87% average mean reduction in TTR and a max reduction of  96%, surpassing that of Patrisiran and other therapies. In our view, as Intellia escalates the therapeutic dose, higher reductions in TTR will lead to better patient outcomes.

We believe this data will enter the history books, signifying that in vivo gene editing and one-time cures are possible. The only question remaining is will it be durable?

Researchers from MIT and the Institut Pasteur in France recently published a powerful new genome assembly algorithm called mdBG. Optimized for long-read sequencing data, genome assembly algorithms are one of the most accurate methods for analyzing genetic variations.

Despite its potential, genome assembly’s computationally-intensive process has limited its use. Strikingly, mdBG produces high-quality genome assemblies using 350X fewer CPU-hours and 20X lower peak memory than other state-of-the-art assemblers. Based on our calculations, researchers will be able to assemble a human genome using HiFi sequencing data in just 10 minutes on an Apple M1 laptop computer with only 8 CPU-Cores, as opposed to 60+ hours on an M1 laptop or resorting to expensive, cloud-based compute.

In our view, not only will it make genome assembly with long-read data more accessible but mdBG also could have significant clinical impact. ARK previously has written about the power of long-read sequencing (LRS) in discerning the genes associated with rare disease. With mdBG, LRS-based diagnostics could become faster, more comprehensive, and more reliable than those based on short-read sequencing.

During its annual Sessions event last week, Stripe announced Stripe Reader M2, its first point-of-sale (POS) hardware device designed in-house. Now powering payments for 75% of the world’s top online marketplaces, Stripe previously introduced Stripe Terminal in 2018 for in-store-payments powered by partnerships with BBPOS and Verifone. Terminal includes developer tools (APIs and SDKs), a fleet management solution for customer POS devices, and now the POS device itself. During the past year transactions on Stripe Terminal have soared more than 250% as in-store commerce shut down, according to Kate Brennan, Stripe Terminal Product Lead and ex-Head of Product for Square’s POS business.

Why did Stripe launch its own hardware and how does it compare with that from other POS providers like Square? While Brennan mentions the demand for more elegant hardware designs, Reader M2’s proprietary POS hardware also fits nicely into Stripe’s strategy of verticalizing payment products and services, enabling customers to integrate them with only a few lines of code.

We believe this developer-first approach separates Stripe Terminal from other POS providers such as Square’s POS or FISERV’s Clover. Although BBPOS’s third-party hardware device can integrate third-party payments applications, Stripe Terminal’s other third-party hardware device, Verifone V400, must be programmed. For now, Terminal caters to Stripe’s online enterprise customers who are expanding to customized offline payments, while Square and other POS products offer more turnkey software services focused on small and medium sized businesses. Square does offer a developer platform for more customization but, for now, Stripe Terminal and Square largely solve for different use cases.

On Thursday President Nayib Bukele unveiled details on El Salvador’s plans to adopt bitcoin as legal tender. In the address, President Bukele announced Chivo, a mobile app that will serve as El Salvador’s official Bitcoin wallet. Users will be able to hold both USD and BTC balances in Chivo, exchanging them seamlessly at any time.

To promote the use of bitcoin and encourage citizens to download Chivo, El Salvador will airdrop bitcoin worth $30 to adult citizens who have downloaded the app and registered their wallets. With a population of roughly 6.5 million, 60% of whom are adults, it appears El Salvador will spend up to $117 million to launch this initiative. Cell phone carriers will offer free access to Chivo whether or not mobile phone owners have access to the internet.

Pending the bill’s approval, Chivo will launch officially in September. Please see President Nayib’s remarks on the topic here.

At the Conference on Computer Vision and Pattern Recognition, Andrej Karpathy, Senior Director of Artificial Intelligence (AI) on the Autopilot team, recently spoke about Tesla’s decision to drop radar from its perception system. He reiterated what Elon tweeted not too long ago: camera-based vision is more accurate than radar, and radar introduces more noise than helpful data in a sensor fusion system. Karpathy compared the performance of vision-only with sensor fusion when a vehicle moves toward and then under a bridge, concluding that the superior vertical resolution of the vision-only version of FSD (Full Self Driving) helped distinguish the bridge from a stationary object that would stop the car.

Why, unlike Tesla, have most other companies opted to include data from inexpensive radar and LiDAR in the autonomous solutions space? One argument could be that its 1.3-million-unit vehicle fleet gives Tesla the ability to train cars with 6,000 hand-picked clips of a new release. In other words, Tesla may have “earned” the ability to remove radar from Autopilot thanks to the size of its fleet, a significant advantage we believe relative to its autonomous competitors.

Tesla recently unveiled details on an in-house supercomputer that will train artificial intelligence (AI) and enable autonomous driving. Estimated to be the fifth largest globally, Tesla’s supercomputer is powered by a vast array of Nvidia GPUs and volumes of ultra-fast storage totaling 1.8 exaFLOPS to churn through billions of miles of real-world driving data. Clearly, Tesla is aiming to produce AI models that will meet and exceed human-level driving abilities.

Along with its successor, Dojo, which will “take this to the next level” according to Telsa’s head of AI, this supercomputer could reduce the time and cost necessary to train models so that AI experts can experiment and tune them more quickly. Once ready, the models will be deployed on custom-designed chips running inside each vehicle, fulfilling Tesla’s strategy of vertically integrating its AI capabilities.