Medical AI

## AI

AI Models

• ChatGPT

• Google Gemini

• Anthropic Claude

• Microsoft Copilot

• xAI Grok

• IBM Watson

AI Image Generation

• ChatGPT Sora

• Google Gemini Nano Banana

• Midjourney

Browsers

• Perplexity Comet

• OpenAI Atlas

• Google Disco

Developer Tools

• Claude Code

• Cursor

• Amazon Kiro

• Google Antigravity

Language Translators

• Google Translate

• Meta Omni lingual

Medical AI Models

• ChatGPT Rosalind

• Anthropic Claude Science

• Google Coscientist

• Microsoft Quine

• IBM MAMMAL

• Open Evidence

Cybersecurity

• Anthropic Mythos

• Open AI GPT 5.4 Cyber

AI Agents

• Meta Muse

• Open AI Dots

## Computing

World Simulation Model

• Google World Simulation Model

• NVIDIA Cosmos Omniverse

• OpenAI Sora Physical AI

• Tesla’s FSD and Optimus are leading in embodied physics. Tesla uses a unified neural world simulator

• World Labs Atlas - Marble

AI is being used to optimize data centers and design next-generation AI chips, accelerating its own development.

• Project AlphaChip

• Model Context Protocol

• Model Hardware Standard 

## Medical AI

Digital Biology - Nvidia BioNemo, GPT Rosalind

Digital Chemistry - Chemcrow

DNA Sequencing - PacBio Uses NVIDIA Tensor Core GPUs in its Revio system, which is designed to sequence human genomes at scale.

Drug Discovery - Nvidia BioNemo

Medical AI - Google Med Palm 2

Medical Imaging

Protein Folding - Google Alpha Fold 3, Google Alpha Proteo

Alpha Genome maps 9 billion dna variants

RNA 3D structures - NuFold

Surgery - MIRA surgical robotic platform 

## Drug Discovery

Medical AI Models

• ChatGPT Rosalind

• Anthropic Claude Science

• Google Coscientist

• Microsoft Quine

• IBM MAMMAL

• Open Evidence

- AI has accelerated drug discovery by rapidly screening millions of compounds and predicting their therapeutic potential, leading to the identification of over a million potential antibiotics.

DNA, RNA, Proteins 

DNA Sequencing - PacBio DNA sequence machine

Metagraph - MetaGraph compresses and indexes these gargantuan public datasets, allowing researchers to plug in a DNA sequence and instantly track where else it exists in nature

AlphaGenome - 9 billion dna mutations. Designed to solve sequence interpretation. While MetaGraph tells you where a sequence is, AlphaGenome tells you what it does. It is a foundational machine learning model trained to look at a DNA sequence and forecast complex molecular processes—such as how much RNA it will produce, where it will splice, and how proteins will bind to it.

Prime editing is a "search-and-replace" genome-editing technology developed in 2019 by Dr. David Liu's laboratory that allows for targeted insertions, deletions, and all 12 types of base-to-base conversions without creating double-strand DNA breaks.

Evo 2: A monumental genomic foundation model trained on over 9.3 trillion nucleotides across 128,000+ species. It functions at single-nucleotide resolution to read, predict, and generate genetic sequences from scratch, effectively mapping the evolutionary language of DNA and RNA. 

RNA

CRISPR, TIGR, and VIPR represent distinct generations and lineages of RNA-guided genetic systems that are reshaping the landscape of biotechnology and genome engineering. While CRISPR remains the gold standard, the discovery of TIGR and VIPR introduces smaller, more flexible, and evolutionary ancient alternatives that bypass some of CRISPR's strict biological limitations.

1. CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)

The Baseline: Discovered as an adaptive immune system in bacteria, CRISPR uses a continuous single-guide RNA to match an invader's DNA "mugshot" and cut it.

The Limitations: CRISPR proteins (like Cas9) are very large, making them hard to package into the viral vectors used in gene therapies. Furthermore, they require a specific neighboring DNA motif called a PAM sequence to function, restricting where they can edit.

2. TIGR (Tandem Interspaced Guide RNA)The Next Generation: Discovered by Feng Zhang’s lab at MIT/Broad Institute, TIGR-Tas systems function completely independently of traditional CRISPR pathways.

Why It's Different: Instead of reading one strand of DNA, its tigRNA uses a dual-guide system to latch onto both strands of the double helix simultaneously.

The Advantage: TIGR does not require a PAM sequence, meaning it can theoretically be programmed to target any site across a genome. Because the associated Tas proteins are incredibly compact, engineered variants like TIGRa fit easily inside standard therapeutic delivery vehicles.

3. VIPR (Viral Interference Programmable Repeat)The Ancient Progenitor: Discovered by Jennifer Doudna’s lab at UC Berkeley, VIPR is an ultra-ancient system (potentially 4+ billion years old) found inside viruses. It rewrites the history of CRISPR, suggesting bacteria actually "stole" the foundation of their immune system from viruses.

Why It's Different: VIPR completely breaks the rules of continuous genetic reading. It uses a gapped, noncontiguous code—its guide RNA reads two letters, skips one, and reads the next two. Structurally, it doesn't pry DNA apart; it forms a geometric triplex by snaking entirely around the outside of the helix.

The Advantage: The "skip-one" rule allows it to successfully bind to target viral sequences even if the target mutates its "wobble" positions. It is the smallest RNA-guided system ever discovered, making it an extraordinary candidate for minimal-footprint gene silencing, epigenetic modification, and DNA imaging.

Proteins 

AlphaFold3 - 200 million proteins

AlphaProteo - designed to create novel, high-strength protein binders.

IsoDDE - Hailed as an "AlphaFold 4–level" breakthrough, this highly powerful, proprietary diffusion model generates drug molecules with near-perfect binding accuracy. It operates completely closed-source, drastically outperforming legacy physics-based simulation methods by accurately predicting drug-protein binding and antibody interactions without needing experimental crystal structures.

MitoMatch is a newly developed artificial intelligence framework in molecular biology that maps the entire human mitochondrial proteome. It provides a comprehensive compendium of 2,895 predicted protein-protein interactions within human mitochondria.

Universal Virtual Cell

Biohub’s effort to build a “Universal Virtual Cell” that could let scientists test biological experiments virtually before conducting them in the lab. We envision a virtual cell as a multi-modal, multi-scale, dynamic, and stateful computational system capable of simulating the activity and behavior of a living cell. Scaling datasets toward a universal virtual cell model over the next five years.

Bodyoids

• Brain Organoids

• Retina Organoids

• Liver Organoids

• Kidney Organoids

• Stomach Organoids

• Intestinal Organoids

• Heart Organoids

• Pancreas Organoid

• Bladder Organoid

• Nortochord

Stem Cells - IPSC

• Neural Stem Cells

• Retinal Stem Cells

• Blood Stem Cells

• Bone Stem Cells

• Muscle Stem Cells

• Nerve stem cells

• Skin stem cells

• Heart stem cells

• Spine Tissue stem cells

• MSC - Umbilical Cord

• Bone Marrow

• Fat cells

• Immune cells

• Epithelial cells

Fertilization cells 

• Egg cells

• Sperm cells

An adult human body contains approximately 30 to 37 trillion cells, with estimates varying by sex and age: 

• Adult Male: ~36 trillion cells

• Adult Female: ~28 trillion cells

• Child: ~17 trillion cells 

Distinct Cell Types

• Number of Types: Scientists estimate there are around 200 to over 400 distinct cell types, depending on how broadly or narrowly cell classes are categorized (The Cell Ontology catalogs over 2,500 functional/structural classes)

## Science

1. Anthropic: The Structural Workflow Approach

Instead of engineering specialized biological weights, Anthropic focuses on workspace integration and natural language interfaces for scientists. 

Claude Science: A specialized AI workbench application for life sciences. Rather than deploying a new model, it layers general-purpose Claude models over 60+ database connectors. It seamlessly plugs into laboratory ecosystems like Benchling, BioRender, and 10x Genomics, automating heavy regulatory writing, literature synthesis, and clinical study report drafting. 

2. OpenAI: The Reasoning-First Strategy

OpenAI focuses on deep scientific reasoning capabilities, leaning on inference and specialized training.GPT Rosalind: A specialized family of large language models fine-tuned specifically for biological reasoning and scientific discovery. Named after DNA pioneer Rosalind Franklin, this model is designed to tackle advanced hypothesis generation, biochemical reasoning, and complex cross-disciplinary scientific problem-solving rather than just data retrieval. 

3. Google DeepMind & Isomorphic Labs: The Closed & Open Powerhouses

Google dominates the computational biology frontier with a split strategy: highly specialized, open-source niche models, and a massive proprietary drug discovery engine. 

Google Co-Scientist: A sophisticated, compound multi-agent AI system built on Gemini. It mimics a full research team (employing Generation, Reflection, Ranking, and Evolution agents) to generate, debate, and evolve novel research hypotheses from vast amounts of literature. It has successfully compressed years of lab-pacing research (such as antibiotic resistance pathways) into mere days. 

Alpha Genome: An open-source medical research model utilizing a hybrid architecture of convolutional neural networks and transformers to analyze DNA base pairs and predict molecular properties. 

AlphaFold: The Nobel Prize–winning foundation of modern structural biology that successfully mapped the protein universe. 

IsoDDE (Isomorphic Labs Drug Design Engine): Hailed as an "AlphaFold 4–level" breakthrough, this highly powerful, proprietary diffusion model generates drug molecules with near-perfect binding accuracy. It operates completely closed-source, drastically outperforming legacy physics-based simulation methods by accurately predicting drug-protein binding and antibody interactions without needing experimental crystal structures. 

4. NVIDIA: The AI Foundry Infrastructure

NVIDIA acts as the infrastructure bedrock, providing the foundational compute pipelines and massive biological scale models.NVIDIA BioNeMo: A robust, agentic cloud toolkit and enterprise platform optimized for running heavy biomolecular models (like AlphaFold and OpenFold). It functions as the backend that other tools (including Claude Science workflows) hook into to run high-throughput molecular screening. 

Evo 2: A monumental genomic foundation model trained on over 9.3 trillion nucleotides across 128,000+ species. It functions at single-nucleotide resolution to read, predict, and generate genetic sequences from scratch, effectively mapping the evolutionary language of DNA and RNA. 

Boltz-2: Part of the rapidly accelerating open-science ecosystem pushed by computing partners, offering high-fidelity structural predictions to rival proprietary biology frameworks.

5. IBM: The Multimodal Unifier

IBM MAMMAL (Molecular Aligned Multi-Modal Architecture and Language): IBM's breakthrough, open-source foundation model trained on over 2 billion biological samples. Unlike models that only look at one biological facet, MAMMAL natively unifies protein sequences, small molecules, antibodies, and gene expression profiles into a single computational language. It famously upended oncology consensus by successfully predicting that a specific blood cancer drug would be highly effective against solid tumors—a claim later verified in the lab.

Cancer Research Over 600 New Cancer Models Unveiled After 10-Year Project With Help From 2,800 Patients – And They're Publicly Available Too. Scientists have nearly doubled the number of cancer models available for future research. All these models generated through the Human Cancer Models Initiative (HCMI) have been stored at the American Type Culture Collection (ATCC).