Technical Overview for Pharmaceutical R&D

EEG Foundational
Brain Model

A large-scale AI model trained on heterogeneous EEG data to enable robust biomarkers for CNS drug development and clinical diagnostics.

150K+
Subjects Pretrained
80%+
Cross-site Accuracy
19+ Channels
1024-dim Embeddings
Self-supervised Learning
Scroll to explore
The Challenge

Why Traditional Approaches Fail

CNS research faces a fundamental data problem that standard machine learning cannot solve.

Small Sample Sizes

CNS trials typically have only 100–500 subjects — insufficient for training robust AI models from scratch.

Noisy, Heterogeneous Data

EEG data varies across sites, devices, and settings. Different sampling rates, channel configurations, and protocols.

No Ground Truth

Complex brain disorders lack clear 'ground truth' labels. Disease heterogeneity remains invisible in clinical practice.

R&D Inefficiency

Drug development stalls without proper stratification tools. Small trials can't generate reliable biomarkers.

The Core Limitation

Training AI models from scratch on each trial dataset is impossible. You need the model to already understand the "basic rules" of EEG before focusing on disease-specific signals.

100-500 subjects per trial
Not enough for deep learning
~100
Typical Trial Size
0
Ground Truth Labels
Site Variations
The Solution: Foundational Model Pretraining
Data Pipeline

From Raw EEG to Universal Embeddings

The normalization layer handles variable channel counts and sampling rates, pooling EEG from multiple environments into a single pretraining corpus.

Raw EEG Input

Resting-state EEG recordings from multiple sites, devices, and protocols. Up to 10 minutes per session.

Variable channel counts (19+ electrodes)
Different sampling rates across devices
Multiple clinical and research environments
Multiple Sites
Various Devices
Different Protocols
Unified Embedding Space
Model Architecture

Why EEG Is Not Just Another LLM

The transformer architecture is inspired by LLMs, but custom-built for EEG's unique properties: continuous signals, no discrete tokens, and multi-resolution structure in time, frequency, and space.

Continuous Signal

No natural pauses or discrete tokens like words

No Semantics

Waveforms don't have explicit semantic meanings

Multi-Resolution

Structure in time, frequency, AND space simultaneously

Dual Attention Mechanism

The model attends across both time and channels to capture network-level brain patterns.

t=0s
t=1s
t=2s
t=3s
t=4s
t=5s
t=6s
t=7s

Temporal Attention

The model attends to past and future 1-second segments within each channel, learning patterns over time. This captures how brain activity evolves and helps distinguish signal from noise.

Transformer Stack

Multiple transformer layers process the embeddings, with a decoder that reconstructs EEG segments for self-supervised training.

Embedding Layer
Multi-Head Attention
Feed-Forward Network
Decoder/Reporter
Layer 1
Layer 2
Layer 3
Layer 4
Self-Supervised Training

Learning Without Labels

The model learns EEG structure through reconstruction tasks — predicting masked segments and forecasting future activity. No labeled data required for pretraining.

Training Objectives

Why Self-Supervised?

No ground truth labels exist for complex brain disorders. Self-supervision lets us learn from the structure of EEG itself.

Masked Segment Reconstruction
?
?
?
t=0sContext feeds reconstructiont=10s

Learns Signal vs Noise

Forces the model to distinguish meaningful patterns from random fluctuations.

Captures Dependencies

Learns temporal and spatial dependencies across the entire recording.

👥
150K+
Subjects
⏱️
10 min
Per Session
📊
1024
Embedding Dim
🧠
Resting
State Only
Fine-Tuning

Adapting to Clinical Tasks

Once pretrained, the model is fine-tuned for specific endpoints using relatively small labeled datasets. The foundational knowledge transfers.

Downstream Tasks

Alzheimer's Disease Marker

Case Study: Diagnostic Biomarker

~1,200
Patients
3-way
Split
~400
Per Set
1
Fine-tune on ~400 patients
2
Test on separate ~400 patients
3
Validate on final ~400 patients
Avoids data leakage, ensures generalization

Rule of thumb: Out-of-sample accuracies above ~80% on external tasks indicate the model has captured the main EEG signals needed for clinical work.

This is a heuristic for internal model selection, not a performance claim. Task-specific results are reported with their cohort and comparator on our evidence page — including placebo-response prediction at 69% balanced accuracy across three independent datasets (Neuroinformatics, 2025). Multi-site EEG is prone to site leakage; we control for it with leakage-aware splits (IEEE ISBI, 2023).

Validation & Generalization

Built for Real-World Robustness

Any biomarker or diagnostic must work across sites and populations, spanning multiple centers and geographies — not just on a single dataset.

Multi-Site Testing

Validation across diverse clinical environments

Different EEG devices
Multiple countries
Various clinical protocols

Data Leakage Prevention

Rigorous separation of training and test data

Three-way data splits
Independent validation sets
Never-seen-before testing

Clinical Relevance

Focus on real-world pharmaceutical needs

Diagnostic biomarkers
Treatment response prediction
Placebo response modeling

Cross-Site Generalization

Biomarkers must work across sites and populations — not just on a single, curated dataset.

🇺🇸
Site A
Device X
🇩🇪
Site B
Device Y
🇯🇵
Site C
Device Z
Validation Protocol

All testing and validation are performed on data the model has never seen before, collected with different devices, clinicians, and often in different countries.

Never seen
Test data
Multi-site
Validation
Real-world
Robustness
Pharma Applications

How Partners Use the Model

Partners start from a concrete question — indication, mechanism, or trial design — and the model is fine-tuned and validated for that specific target.

Indication

Alzheimer's Disease

Early detection and progression monitoring through EEG biomarkers

Key Benefits
Objective diagnostic support
Treatment response tracking
Clinical trial stratification

Partner Engagement Flow

Define Target
Indication, MoA, or trial
Apply Model
Pretrained on 150K+ subjects
Fine-tune
Disease-specific signal
Validate
Cross-site robustness

The Foundational Advantage

Large-scale pretraining + rigorous cross-site validation + targeted fine-tuning enables robust biomarkers that generalize beyond a single study.

Reduce Noise
Filter site differences
Focus Learning
On disease-specific signal
Enable Biomarkers
That truly generalize