This week in TechBio 2024/03/11
This week saw the release of RoseTTAFold All-Atom paper, as well as the associated code. There were also startup headlines in the AI x medtech space (PreemtiveAI, Overjet, and Roche). Insilico Medicine also published a paper about its first AI-discovered drug.
Pharma and Biotechs
2024/03/10 New York Times: Coverage of recent biological foundation models (specifically UCE, GeneFormer, and scGPT). While it is cool to see computational biology research being presented in the mainstream press, the journalists clearly overhype the technology. Indeed they take the example of Norm cells that took 130 years to discover (starting the clock before we even knew the structure of DNA), which could be “discovered” by embeddings of large cell atlases. It however completely misses the point that these models were trained on the very data that was used to discover this biology; deep learning models are not succeeding against biologists, but instead thanks to their work crafting well designed panels of data. It also ignores that “biologists” have long since adopted modern analysis technique, and that Norm cells were identified using the very same technology (single-cell RNA-seq followed by dimension reduction and clustering) that these foundation models rely on.
2024/03/08 Responsible AI x Biodesign: A large part of the protein design community signed a commitment to responsible use of protein and molecular design AI tools. It kinda feels like a promise to be extra good boys/girls and doesn’t really have any specific or binding terms.
2024/03/08 STAT: Amit Etkin, founder and CEO of Alto Neuroscience, writes an interesting piece arguing for “precision psychiatry”. He posits that the current way psychiatry is conducted, trying drugs for a given set of symptoms until one works, is due to the lack of good biomarkers for diagnostics. He further argues that this lack of biomarkers (or anything associated with the cause) further limits approval of new drugs; indeed a drug that works really well for a small population will look bad on a random sample of the population. Having better biomarkers (via EEG or cognitive tests) could help with patient stratification, thus improving clinical trials success rates and patients outcome alike.
2024/03/07 Fierce Biotech: Roche introduces a new predictive model for continuous glucose monitoring. This tool is used to predict glycemic levels in the 30-120 minutes range, which could be useful for people with diabetes. I have to be honest and say that it does not sound like a very long timeframe. Indeed weather is much harder to forecast, yet we can forecast it on a much larger timeframe. They plan on improving that later with better models, or use a time series pretrained model (e.g. TimesFM).
2024/03/07 STAT: A recent study shows that Beyfortus, an RSV monoclonal antibody, achieved an effectiveness of 90% against RSV-associated hospitalization in infants in their first RSV season. Despite a $495 price point, the supply was low enough to require some triage and to only give it to the youngest children.
2024/03/06 STAT: Cerner, now part of Oracle Health, currently holds about a quarter of the EHR market in the US compared to Epic’s greater than one third. It however announced that it plans on developing GenAI tools (competing against Abridge and MSFT Nuance) as well as analytics that would work in an EHR-agnostic fashion. They would thus be able to get a lot of value from the hospital software market, even if they lost the fight on the tools used to store that data. They even allow themselves to be a bit cheeky, saying that Epic runs on Oracle’s databases anyway.
2024/03/06 Yahoo!Finance: PreemptiveAI, a company designing models for wearable devices and Photoplethysmograms (PPG), comes out of stealth with a $6.4m seed deal. Their website and press release are however extremely (quite frankly excessively) boastful, to the point of silliness.
2024/03/05 Kaleidoscope: an interesting blogpost pushing the thesis that in the bio and medical fields, all data are not created equal and are not equivalent. A simple example would be a disease prediction model trained only on adults, which would most likely not generalize for children. This also joins the answer of GSK’s head of ML gave to IsomorphicLabs last week. However, do note that this post is written by a company selling tools for building and searching datasets so they may be biased.
2024/03/05 Fierce Biotech: Alnylam’s RNAi therapy succeeded in its phase 2 trial for hypertension, inching closer to the potential $2800m milestone payment from Roche.
2024/03/05 Fierce Biotech: Overjet, a company that designs AI for dentistry X-ray, raises $153m series C. This announcement comes right after their cavity detection software got an approval from the FDA.
2024/02/28 Fierce Biotech: Medidata, a Dassault System subsidiary, and Thermo Fisher extend their collaboration on building AI technology for improving enrollment in clinical trials.
Papers:
2024/03/08 Nature Biotech: “A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models”. Paper about InSilico Medicine’s first AI-designed drug.
2024/03/07 Science: “Generalized biomolecular modeling and design with RoseTTAFold All-Atom”. RoseTTAFold All-Atoms, and RFDiffusion All-Atoms, as well as the associated code (here and here) are finally publicly available.
2024/02/29 Nature Communications: “Drug target prediction through deep learning functional representation of gene signatures“. Novartis presents “Functional Representation of Gene Signatures (FRoGS)“, a way to obtain semantic gene embeddings (like gene2vec).
BiB events:
Bits In Bio just started a Shanghai chapter, come and visit them !
2024/03/12: “RTP - Personalized cancer vaccines: a very cool idea that doesn’t quite work yet” — Biolabs NC, 701 W Main St #200, Durham, NC 27701, USA
2024/03/12: “Bits in Bio Happy Hour Sponsored By Ganymede” — Hilton Boston Park Plaza, 50 Park Plaza, Boston, MA 02116, USA
2024/03/13: “Bits in Bio x Nucleate x Valence Labs” — https://lu.ma/ixohxxng
I very likely missed quite a few announcements, don’t hesitate to DM me @gama_search if you see anything missing or needing corrections. Or DM me if you have seen some cool paper / news coverage that you would like to see on next week’s newsletter.