#285: The Federal Aviation Administration Has Grounded and Is Investigating Virgin Galactic, & More

The FAA is investigating Virgin Galactic after learning about a mishap during the flight that carried Richard Branson to space. While the flight looked picture-perfect, the spacecraft deviated from its planned trajectory, potentially putting its landing at risk. In a scathing report, the New Yorker highlighted Virgin Galactic’s lackadaisical processes and procedures, noting that it fired a flight-test director who raised concerns about the space program’s safety.

The New Yorker also contrasted Virgin Galactic’s strategy and technology to those of its competitors, Blue Origin and SpaceX. At Virgin Galactic, humans control extraordinarily complex systems, while Blue Origin and SpaceX have automated them. ARK believes that automation adds to the safety and security of complex systems, not just rockets but all vehicles.

Last week, scientists at Google (GOOGL) and Pacific Biosciences (PacBio, PACB) published a deep learning algorithm called DeepConsensus (v0.1) that improves PacBio’s high-fidelity (HiFi) sequencing data for both research and clinical applications. Notably, DeepConsensus incorporates machine learning principles adapted from natural language processing (NLP).

Since even single-letter mistakes can cause a misdiagnosis, downstream research and clinical bioinformatics pipelines require inputs of high-quality sequencing data. DNA sequencing itself is error prone. To minimize errors, sequencers read the same DNA letters many times over, averaging the signals into a consensus—an agreement.

DeepConsensus operates alongside a PacBio algorithm (pbcss) during the first step of DNA sequence processing called primary analysis. The model treats streams of sequenced DNA letters much like a speech recognition algorithm interprets meaning in a sentence. An NLP model, for example, might focus its attention in a sentence on nouns to assign meaning to an ambiguous, mid-sentence word like “it”. Similarly, DeepConsensus compares small sections of sequence reads along with the associated metadata* to eliminate sequencing errors during the generation of data.

Eliminating errors, DeepConsensus generates faster, more cost-effective, and more accurate HiFi reads. In our view, their deep understanding of HiFi’s weaknesses was key to the researchers’ success. Specifically, the authors trained DeepConsensus with a “gap-aware” method that accounted for systematic “gaps” in HiFi sequencing data that occur near repetitive sections of DNA.

In future DeepConsensus iterations, the teams aim to accelerate the algorithm and enable broad-based adoption. Future long-read sequencing systems are likely to feature onboard GPUs for the acceleration of AI inference.

According to the authors, DeepConsensus heuristics also might boost Oxford Nanopore Duplex long reads as well as algorithms used for decoupling mixed DNA samples batched together during a sequencing run. Both these applications involve aligning many similar sequences together to look for small differences. We look forward to seeing how scientists apply DeepConsensus principles to other challenges in biology.

*Subread partitions are tensor objects that include base-calls, pulse widths, interpulse durations, signal-to-noise ratios, and strand identities.

Last week, Gary Gensler stated in an interview that the SEC is considering banning payment-for-order-flow (PFOF). Many commission-free retail brokers like Robinhood, Schwab, and Interactive Brokers route trade orders to market makers in exchange for compensation on a per-share basis. Market makers then take their piece of the action from the spread, the difference between bid and ask, and add to the liquidity of the overall market.

US regulators have criticized PFOF in the past, with concerns that retail brokers will route orders to maximize their compensation instead of optimizing trade execution for retail investors. In recent comments, Gensler added that he believes market makers access trading data not available to other investors.

Today, retail brokers must disclose their PFOF publicly. Robinhood, for example, receives roughly $0.23 per 100 equity shares and $0.60 per options contract, and it executes 95% of trades equal to or better than the best available exchange price, or NBBO (National Best Bid and Offer). In comparison, Schwab executes 92.4% of its trades in S&P 500 stocks at prices equal to or better than NBBO.

In ARK’s view, on balance PFOF results in lower prices and better execution for retail investors. Alternatives like commissions have not worked well. That said, we always will welcome the SEC’s efforts to protect investors, hoping they will continue to encourage innovation and competition, ultimately serving retail investors well.

In August non-fungible token (NFT) trading volume surged as decentralized marketplace OpenSea crossed a record-breaking $3 billion in monthly volume, up more than 12-fold from $248 million during July, its second-best month.

At this time, three NFT categories are driving this explosion in growth:

• One-of-One Pieces: unique creations that can span multiple mediums including digital and generative artwork, music, and 3D renderings
• Avatars: one-of-N collections such as CryptoPunks and Bored Ape Yacht Club which purchasers use to represent their digital personas in social media profile pictures
• In-Game Assets: NFT-based items that purchasers can trade across platforms in play-to-earn games like Axie Infinity, as ARK featured last month.

Despite the staggering growth in trading volume, many investors find the intrinsic value of NFTs difficult to grasp. The launch of a new collection dubbed “Loot” illustrates the challenge. Created by Vine co-founder Dom Hofmann, the project includes 8,000 NFTs, each of which contains a plain-text list of “randomized adventurer gear” but no images, no stats, no rules, no roadmap of any kind. Instead, the community claiming the NFTs will create their own use cases for the NFTs in a build-your-own-adventure game. In little more than a week since its launch, Loot spurred a frenzy of products built around the original text, generating more than $200 million in volume on OpenSea.

In addition to the rise of play-to-earn gaming and the democratization of art, Loot illustrates how NFTs are enabling new forms of social collaboration and community engagement, all centered around digital asset ownership. Already, brands ranging from Budweiser to the US Open are experimenting with this new technology to increase their customer engagement.