Cancer Testing on a PostCard

Plus: Cancer Vaccines are Working

April 27, 2024

Hello Tensor Black Fans!

What if cancer testing were as simple as… mailing a postcard?

Learn about a dried bloodspot AI that detects three kinds of cancer, currently used for population health studies in China.

Level up this week:

  • Cancer Vaccines are Working

  • $1B for GenAI Drug Discovery

  • Detecting Cancer from Postcards

  • China Unveils Powerful Video AI

  • AI Designs CRISPR Editors

Doctors have begun trialling hundreds of patients in the world’s first personalized mRNA cancer vaccine for melanoma, as experts hailed its “gamechanging” potential to permanently cure cancer.

Details

  • One needle samples a patient’s tumor

  • DNA of the tumor is read by next generation sequencing, revealing the code for antigens — markers on the surface of cancer cells

  • The custom code is inserted into a custom mRNA nano lipid particle, personalizing a vaccine.

  • A new, personal vaccine is produced and tested

  • After successful tests, the patient is injected to activate their immune system to attack the disease.

Why it Matters

Personal vaccines powered by nanoscale data rely on “platform approvals” from the FDA and similar agencies.

Successful trials like the one in England show promise for a future where many melanomas (cancers with markers on their surface) can be managed with “mRNA platform” drugs.

Xaria Therapeutics raised $1B to chase generative AI for rapidly accelerating drug discovery.

Why it Matters

Early academic results in 2022 identified a new antibiotic in 48 minutes, or drug candidates even faster at Isomorphic Labs.

David Baker’s lab won a “Breakthrough Prize” in 2021 for combining alpha fold with molecular modeling to guess how a protein interacts with DNA, RNA and antigens.

Arch Venture’s largest bet on drug discovery will help mainstream these promising techniques.

Less than 0.05 milliliters of dried blood on a postcard could be used to detect gastric, colorectal and pancreatic cancer, early research suggests.

Details

  • 17 dried serum spots (DSS) of blood are sent to a laboratory on a postcard

  • Reagents are added to emphasize levels of glucose and lactic acid

  • Mass spectrometers plot glucose and lactate concentrations

  • A machine learning (ML) model identifies peak patterns in glucose and lactate indicative of gastric, colorectal and pancreatic cancer.

Why it Matters

Cost is a major inhibitor for widespread use of phenomics, where extracting nanoscale data or safely transporting biological samples can be expensive.

This research enables testing for a few dollars with remarkably inexpensive and convenient transportation.

Dried blood samples could one day help propel AI-driven cancer diagnosis to mass adoption.

China's text-to-video model Vidu was unveiled in Beijing on Saturday April 27th, intensifying the artificial intelligence competition globally.

Details

  • Video is the ultimate “multi-modal longitudinal record.”

  • Video, sound, text captions, talk tracks, thumbnails and more are aligned over time.

  • An AI is built to understand each component, then generate soundtracks, captions, thumbnails and more.

  • Vidu challenges Sora from OpenAI and VideoPoet from Google DeepMind.

Why it Matters

Phenomics relies on understanding multi-modal biological data synced over time. The science and technology behind video models is one key to understanding life at a nanoscale.

Profluent announced the release of OpenCRISPR-1, an open source AI that creates CRISPR gene editors.

Details

  • CRISPR needs “editing genes” like Cas9 that align with the human genome and insert changes.

  • Profluent trained LLMs on a massive dataset of CRISPR operons culled from hundreds of studies (26 terabases).

  • The LLMs then generated de novo gene editors not seen in nature, but capable of editing different parts of the human genome.

  • The result is a variant of Cas-9 that won a Nobel Prize, but with hundreds of mutations (see this protocol).

  • CEO Ali Madani predicts a future where “AI precisely designs gene editors for bespoke cures to disease” (see this thread)

Why it Matters

Once we identify genetic variants that lead to disease and corrections that can reverse their course, we need a tool to repair our DNA or RNA.

These gene editors are difficult to design, as you must figure out what sequence of DNA, RNA, and at what length, will be accepted by the human body and affect change.

AI can now accelerate that process, radically reducing the search space and time to develop CRISPR therapies.