We started collecting applications for SciFounder Fellowship!! For mission-driven hard tech founders Up to $1M to get started Direct mentorship from us for 1 year+ Funding decisions within weeks https://t.co/NLt338cqAQ Deadline is 2/28 - 11:59 pm PST

Embryo editing has incredible potential and could eliminate some of the most devastating genetic diseases, but in the wrong hands do more harm than good. So glad that Lucas stepped up and started Preventive to do it the right way.

There's a new kind of biotech company in silicon valley, the 'human enhancement company'. I believe this is the next new category of technology company to yield not only a >$100B household name, but to truly uplevel human health and wellbeing. These companies are

Thanks, @deepchecksvc team, for featuring me as an investor of the week!! It's been a lot of fun working with you -- love the platform you have built for all of us. Would any deep tech founder encourage to check out Deep Checks -- it's a super easy way to get in front of a

I'm looking for the "Anduril" of Bio. Anyone know any Bio-Defense startups?

https://t.co/CfDlNOMRHv

The models aren’t “learning physics” -- they’re memorizing binding patterns from training data.  LLMs should be used as a first filter and physics-based docking, or better experimental binding data, as a last-mile validation.

Even after the changes, the models still forced ligands into the original pockets, sometimes with clear clashes, while reporting high confidence scores. These ligands should not bind anymore.

Researchers stress-tested AlphaFold 3, RoseTTAFold-AA, Chai-1, Boltz-1, by sabotaging the binding pocket of proteins or changing the ligand’s charge, making the interactions energetically impossible.

Protein folding models break the laws of physics... ...at least that's what a recent stress test study is showing ⬇️

or junior directors at medium-sized biotechs who are told there are no resources for their ideas

the most interesting things in the biotech world are hidden in the heads of smart associates and principles at vc firms who strongly recommended against an investment decision that was, in the end, made

(New thesis) Going direct with diagnostics. Consumer Tempus businesses. Concept Function Health and Superpower have demonstrated massive consumer willingness to spend on comprehensive health testing, but they don't provide net new insights beyond standard clinical biomarkers.

(New thesis) AI-ification of R&D proteins There’s been an emergence of new papers on computational antibody, peptide, nanobody, and and enzyme design. While these are all relevant for new therapeutics, there’s massive market opportunity in R&D reagents. (This thesis was started

https://t.co/rgTNFi0sFp

Metsera did an obesity speed run: 2022 – Founded Sep 2023 – Acquired Zihipp (UK peptide startup) for ~$34M upfront (steal!) Apr 2024 – Out of stealth with $290M raise Jan 2025 – IPO ~2.7B market cap Today – Pfizer acquires Metsera for ~$4.9B upfront wow

Not surprising, but still sad to see. On the flip side, the biotechs getting started in one of the toughest markets ever will likely do very well. We want to back founders who are building against all odds. Source: Crunchbase

https://t.co/8PVFqRowuD

It doesn’t look at individual diseases in isolation. Co-morbidities greatly influence each other, so it gets much closer to reality. On top of that, it predicts not just what might happen but also when, which makes the output far more actionable than a static risk score.

Built on ~0.4M UK Biobank participants and validated with independent ~1.93M Danish registry records. They could simulate future health trajectories up to ~20 years and could make multi-disease risk estimates comparable to single-disease models.

This seems like a big deal to me. A GPT-style generative model (Delphi-2M) that models lifetime trajectories across >1,000 diseases + death, conditioned on basic risk factors.