Bio-PrecisionAI Health

03/17/2026

🧠 Double Machine Learning (DML): A Practical Guide to Causal AI in Epidemiology

Most machine learning answers: “What predicts an outcome?”

Double Machine Learning (DML)—developed by Victor Chernozhukov and collaborators answers the more important question:

👉 “What actually causes the outcome?”

# # 🔍 What is DML?

DML is a framework that combines:
- 🤖 Machine learning (to model complex patterns)
- 📊 Causal inference (to estimate unbiased effects)

It works by:
✔️ Cross-fitting (avoids overfitting bias)
✔️ Orthogonalization (separates signal from confounding)

👉 Result: reliable causal estimates even in high-dimensional data

# # ⚙️ Core Types of DML (from basic → advanced)

# # # 🔹 LinearDML (Partially Linear)
- Estimates average treatment effect (ATE)
- Simple, fast, interpretable
👉 Best starting point

# # # 🔹 SparseLinearDML
- Designed for high-dimensional data
- Uses regularization (Lasso-style)
👉 Common in genomics & epidemiology

# # # 🔹 KernelDML
- Captures nonlinear relationships
- More flexible than linear models

# # # 🌲 ForestDML (key method)
- Uses random forests
- Estimates heterogeneous treatment effects
👉 Answers:
- Who benefits most?
- How effects vary across populations?

# # # 🌳 CausalForestDML
- Specialized version of ForestDML
- Optimized for causal heterogeneity
👉 Widely used in:
- precision medicine
- policy targeting

# # # 🔹 DRLearner / Doubly Robust DML
- Combines DML + doubly robust estimation
- More stable under model misspecification

# # # 🔹 Multi-treatment (Multi-class DML)
- Handles multiple treatment groups
- Example: drug A vs B vs C

# # # 🔹 Multi-label DML (emerging 🚀)
- Handles multiple simultaneous exposures
- Captures interaction effects
👉 Very relevant for:
- omics data
- biological systems

# # 🧪 Applications

📊 Epidemiology
- Air pollution → health outcomes
- Drug and vaccine effectiveness

🏥 Healthcare / EHR
- Treatment effects from observational data

🧬 Computational Biology
- Gene expression → disease
- RNA features → protein interactions

💰 Economics & Policy
- Education, policy, and intervention impact

# # 🚀 Why DML matters

✅ Controls high-dimensional confounding
✅ Reduces bias from model misspecification
✅ Handles nonlinear relationships
✅ Enables personalized (heterogeneous) effects
✅ Bridges ML ↔ causal inference

# # ⚠️ Challenges

❌ Requires good overlap in data
❌ Needs careful model tuning
❌ Interpretation becomes complex (especially ForestDML)

# # 💡 Key intuition

- LinearDML → “Does it work on average?”
- ForestDML → “Who does it work for?”
- Multi-label DML → “How do multiple factors interact causally?”

# # 🧠 Big picture

DML transforms machine learning from:
➡️ prediction
to
➡️ causal understanding

And that shift is critical for:
- epidemiology
- healthcare
- AI-driven science



~ ChatGPT

03/17/2026

🌍🧠 World Models: Teaching AI to Understand and Imagine Reality

What if an AI didn’t just react to the world, but could build an internal model of it, simulate outcomes, and plan ahead?

That’s the idea behind World Models, introduced by David Ha and Jürgen Schmidhuber in their influential paper World Models.

# # 🔍 What are World Models?

A World Model is an internal representation that an agent learns to:

✔️ Understand how the environment works
✔️ Predict future states
✔️ Simulate possible outcomes
✔️ Plan actions before taking them

👉 Instead of learning only from trial and error, the agent learns to “think before acting.”

# # 🧩 Core Architecture (3 Components)

# # # 1️⃣ V — Vision Model (Perception)
- Encodes raw observations (e.g., images, states)
- Typically a Variational Autoencoder (VAE)

👉 Compresses high-dimensional input into a latent space

# # # 2️⃣ M — Memory Model (Dynamics)
- Predicts how the world evolves over time
- Often implemented with:
- RNNs (e.g., LSTM)
- or modern sequence models

👉 Learns:
zₜ₊₁ = f(zₜ, aₜ)

# # # 3️⃣ C — Controller (Decision Maker)
- Chooses actions based on latent state
- Usually small (even linear in original paper!)

👉 Key idea:
- Complexity is in the world model, not the policy

# # 🔄 Training Pipeline

# # # Step 1: Learn Representation
Train VAE:
x → z

# # # Step 2: Learn Dynamics
Train sequence model:
(zₜ, aₜ) → zₜ₊₁

# # # Step 3: Train Controller
- Use reinforcement learning or evolution strategies
- Train inside the learned simulated world

👉 This is powerful:
- No need for expensive real-world interaction

# # 🧠 Key Idea: “Dreaming”

The agent can:
- Simulate trajectories
- Train entirely in its own imagination

👉 Called:
“Dream environment”

This dramatically improves:
- sample efficiency
- safety
- scalability

# # ⚙️ Modern Extensions

# # # 🔹 Latent World Models
- Work entirely in compressed latent space
- Faster and more scalable

# # # 🔹 Transformer World Models
- Replace RNNs with Transformers
- Capture long-range dependencies

Used in systems like:
- Decision Transformer

# # # 🔹 MuZero-style Models
Developed by DeepMind

- Learn dynamics without explicit environment rules
- Combine planning + learning

Used in:
- MuZero

# # # 🔹 Diffusion / Generative World Models
- Use generative models to simulate environments
- High realism (video, physics, robotics)

# # 🧪 Applications

🎮 Reinforcement Learning
- Game playing
- robotics control

🚗 Autonomous Driving
- Predicting future scenarios
- planning safe actions

🧬 Computational Biology
- Simulating molecular interactions
- protein–RNA dynamics

🤖 Robotics
- Learning from simulation before deployment

# # 🚀 Why World Models matter

✅ Reduce reliance on real-world data
✅ Enable planning & reasoning
✅ Improve sample efficiency
✅ Allow safe training via simulation
✅ Bridge perception + dynamics + control

# # ⚠️ Challenges

❌ Model errors compound over time
❌ Hard to learn accurate dynamics
❌ Latent space interpretability
❌ Scaling to real-world complexity

# # 💡 Deep Insight

World Models shift AI from:

➡️ Reactive systems
to
➡️ Predictive, planning agents

# # 🔮 Big Picture

World Models are a step toward:

- general intelligence
- embodied AI
- agents that can imagine, reason, and act

👉 In essence:

A good world model = the ability to simulate reality before touching it.



~ ChatGPT

03/15/2026

A new OpenFold3-preview-2 release is going public today.

Key highlights:
👉 The current checkpoint, 155K, is competitive with AlphaFold3 and Protenix-v1 on small-molecule runs-and-poses benchmarks.
👉 The full training dataset is being made public, including the self-distillation data and RNA sets. In many model releases, you get weights, inference code, and sometimes a paper, but not the full training data behind the system. Here, the training corpus is accessible as well, which makes it possible to inspect, reproduce, benchmark and modify the model by independent teams.
👉 The release package contains a cleaner distribution path, expanded training documentation, and a clearer public setup for teams that want to evaluate the model themselves.

Open science gives independent teams the ability to test a model under different conditions and in different workflows. That is how trust gets built in this field, and how models improve in the right directions.

We love the OpenFold Consortium for pushing that work forward and contributing to a strong open-source ecosystem for drug discovery. Thrilled to be collaborating with so many fantastic people in the ecosystem.
👏 Mohammed AlQuraishi Mallory Tollefson, Ph. D. Jan Domanski, PhD Woody Sherman Brian Weitzner Christina Taylor Lucas Nivon Peter Clark, PhD Alexandre Zanghellini

The new checkpoint can of course be accessed through ApherisFold, with local inference, result inspection, benchmarking, fine-tuning on in-house data, and API-based integration into existing drug discovery workflows.

🔗 Press release: https://lnkd.in/di8iYgTQ
🔗 Full technical report: chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://lnkd.in/diKSEBM6



Copied

We are enabling validation and rapid iteration so researchers can turn cofolding models into scientific infrastructure that speeds drug discovery

03/14/2026

Met the talent hiring team today at @ Xaira_Thera and they asked me what kind of researchers I want to hire in the age of agentic coding

Honestly took me a second because the answer has changed.

I's not about who codes the fastest anymore, agents handle that.

What I look for: Someone who wakes up obsessed with the right question. Who can smell a bad assumption before it wastes three months of experiments. Who can write a one-page spec clear enough that an agent can execute it — and knows when the agent drifted.

The researchers who stand out aren't the ones who work harder. They're the ones who decide better.

Scientific taste, judgment under uncertainty, the instinct to kill a project early when the signal isn't there.

That's always been what separated great researchers, the difference now is you can't hide behind "I'm still running experiments."😅

If that sounds like you, let's talk.

~ Dr. Wang

03/10/2026

🚨 Yann LeCun, the former chief AI scientist at Meta and a Turing Award winner, just raised $1.03 billion in seed funding for a new startup called Advanced Machine Intelligence Labs, or AMI.

The round values the company at $3.5 billion and is believed to be the largest seed round ever raised by a European company. Investors include Nvidia, Temasek, and Jeff Bezos' Bezos Expeditions.

LeCun has been one of the most vocal critics of the technology dominating the AI industry.

His argument is that large language models, the systems behind ChatGPT, Claude, and Gemini, are fundamentally limited because they only learn from text. They can generate fluent language, but they do not understand the physical world.

AMI is built around a different approach called world models.

Instead of predicting the next word in a sequence, these systems learn from video, audio, and sensor data to build abstract representations of how the real world works. The underlying architecture, called JEPA, is a framework LeCun first proposed in 2022.

The company plans to spend its first year entirely on research and development, with target industries including healthcare, robotics, and manufacturing. The founding team is drawn almost entirely from Meta's AI research organization.

A billion dollars in seed funding for a company openly challenging the direction nearly every major AI lab is moving in.

Most of the industry is betting on making language models bigger. LeCun is arguing the entire approach has a ceiling and that the next generation of AI will need to learn the way humans do, through experience with the physical world rather than text alone.

What are your thoughts on this? 🤔💬

~ Evolving AI

03/06/2026

It was a great time yesterday at our BIO on the Bay event held in St. Petersburg, FL in partnership with NEXI Biotech and others.

Up next is Nucleate Florida’s in event taking place in Gainesville, FL on April 2. This event is organized in partnership with UF Innovate Accelerate and Bio-PrecisionAI Health. Gainesville is a talent hub.

We look forward to bringing together innovators, researchers, investors, entrepreneurs, and industry leaders to explore how AI is transforming biotech and healthcare.

Register here: https://luma.com/mk64kffy

03/02/2026

🚀 Our AI in Biotech Event Is Finally Here! 💃🕺

Mark your calendars!

📅 Date: Thursday, April 2, 2026
⏰ Time: 6:00 – 9:00 PM EST
📍 Venue: UF Innovate, 747 SW 2nd Ave, Gainesville, FL

🚀 As the Partnerships Operations Lead at Nucleate Florida, I’m excited to be leading our efforts to organize AI in Biotech — a collaborative event by Nucleate Florida & UF Innovate, bringing together researchers, founders, AI scientists, investors, and students to explore how artificial intelligence is transforming biotechnology 🤖🧬

Agenda Highlights:
✨ Welcome • AI Keynote + Q&A
✨ Networking & Dinner
✨ Expert Panel • Closing Remarks

Featured Panelists:
• Dr. Scott Siegel – Interim Professor and Postdoc, University of Florida
• Dr. David Vaillancourt – Founder & CSO, Neuropacs
• Dr. Gustavo Seabra – Associate Professor, University of Florida
• Dr. Martin Handfield – President & CEO, Neuropacs
• Dr. Noah Berlow – Co-Founder & CTO, First Ascent Biomedical

Moderators: Joseph Luper Tsenum • Dylan T.

🎟 Register here: https://luma.com/mk64kffy

Let’s build the future of AI-driven biotech together.

02/20/2026

As genomic information becomes increasingly relevant for clinical care and public health, strengthening our collective ability to interpret complex genomic findings matters as much as generating them. That is why I’m excited to share information about Illumina Grand Rounds in Genomic Medicine, a monthly, interactive webinar series modeled after medical grand rounds and molecular tumor boards. This series is designed to help healthcare professionals build practical strategies for interpreting genome-sequence information.

Our second session of the series, which will be led by Illumina oncology Medical Affairs will highlight an ovarian cancer case. Please join us on February 26 at 8:00 AM PT to hear how genomic analyses of BRCA1 and HRD are guiding clinical trials.

Register here: https://lnkd.in/gSPv6YEy

~ Dr. Eric Green

This link will take you to a page that’s not on LinkedIn

02/18/2026

Bio-PrecisionAI Health Transitions to C-Corporation and Strengthens Leadership for Next Phase of Growth

Bio-PrecisionAI Health is entering an exciting new chapter as we continue building toward our long-term mission of transforming drug discovery through AI.

As part of positioning the company for investor readiness and long-term growth, Bio-PrecisionAI Health LLC is officially transitioning to Bio-PrecisionAI Health Inc., converting from a limited liability company (LLC) to a C-corporation. This change aligns with standard U.S. corporate structuring practices and strengthens our foundation for future investment, governance, and scalability.

Alongside this transition, we are announcing an update to our leadership structure:

• Joseph Luper Tsenum, formerly Founder and CEO, will now serve as Co-Founder, CEO, and Director of Research & Development, reflecting his continued leadership in advancing our core AI and molecular design technologies.

• Victoria Harper-Alexander, formerly Chief Scientific Officer (CSO), will now serve as Co-Founder and Chief Scientific Officer (CSO), recognizing her foundational scientific leadership and ongoing role in shaping our research direction.

These changes reflect our evolution as a company and reinforce our commitment to scientific excellence, technical innovation, and responsible growth.

As we continue developing next-generation AI models to design therapeutic molecules across multiple modalities, we are energized for what lies ahead. We look forward to a supercharged 2026 as we scale our platform, expand our capabilities, and move closer to delivering meaningful impact in precision medicine.

01/28/2026

01/26/2026

Layers of AI

#1. Classical AI (Rule-Based AI)
What it is: The earliest form of AI, based on explicit rules written by humans.

components:
- Symbolic AI
- Expert Systems
- Knowledge Representation
- Logic & Reasoning
No learning from data & Rigid and hard to scale

# 2. Machine Learning (ML)
What it is: Systems learn patterns from data instead of fixed rules.

Key components:
Supervised Learning | Unsupervised Learning | Reinforcement Learning
Classification & Regression

#3. Neural Networks
What it is: ML models inspired by the human brain, using layers of neurons.

# 4. Deep Learning
What it is: Neural networks with many layers, capable of high-level feature learning.

Key architectures:
CNNs → images & vision
RNNs / LSTMs → sequences & time series
Transformers → language & multimodal data
Autoencoders → representation learning

Example:
Face recognition, speech-to-text, translation.

#5. Generative AI
What it is: Models that create new content, not just predict.

Key models:
LLMs (text generation)
Diffusion Models (image/video generation)
VAEs
Multimodal Models (text + image + audio)
Example: Chatbots, image generators, code assistants.
Key leap: AI becomes creative, not just analytical.

#6. Agentic AI (Autonomous AI)
What it is: AI systems that plan, remember, decide, use tools, and act autonomously.
Core capabilities:
Memory
Planning
Tool Use (APIs, databases, code ex*****on)
Autonomous Ex*****on

By KODI PRAKASH SENAPATI