Pharma & BioTech
Regulators require uncertainty quantification. Small samples are the norm. Bayesian methods handle both natively.
FDA and EMA require uncertainty quantification. Credible intervals are native to Bayesian output. Small-sample efficiency, interpretability, and the ability to formally incorporate clinical prior beliefs make Bayesian methods the natural fit for pharma and biotech.
We basically have clients in two industries: marketing, and biotech.
Work in This Space
Roche
Large-Scale Hierarchical Bayesian Model
Successfully fit a 34K-parameter model on 250K observations in just over 1 hour using JAX/NumPyro GPU-accelerated inference.
Takeda
CAR-NK Cell Therapy Digital Twin
Bayesian digital twin for a 28-day CAR-NK cell manufacturing pipeline. Stage-by-stage state space model enabling real-time prediction from day-6 measurements.
Akili Interactive
Digital Therapeutics Cognitive Modeling
Bayesian psychometric modeling for EndeavorRx — the first FDA-approved prescription video game for ADHD treatment.
Trusted By
- Roche
- Takeda
- Akili Interactive
- Erisyon
- Haleon
- Syngenta
- IQVIA
- Procter & Gamble
What We Solve
Clinical Trial Analysis
Hierarchical Bayesian designs for dose-response, PK/PD modeling, patient outcomes with longitudinal data, and biomarker estimation with regulatory-defensible uncertainty.
Drug Manufacturing Digital Twins
Cell therapy manufacturing process modeling with real-time prediction from early-stage measurements. Stage-by-stage Bayesian state space models for CQA tracking.
Genomics & Large-Scale Bio Data
Large-scale hierarchical Bayesian models validated at 34K parameters and 250K observations. GPU-accelerated inference via JAX/NumPyro in approximately 1 hour.
Digital Therapeutics
Ordinal regression for clinical assessment instruments. Bayesian psychometric modeling for cognitive constructs and latent variable modeling for digital biomarkers.
Why Bayesian
Why Bayesian for Pharma & BioTech?
- Regulatory defensibility — FDA/EMA require uncertainty quantification. Credible intervals are native to Bayesian output. Models pass regulatory validation where black-box ML fails.
- Small sample efficiency — Hierarchical models work with as few as 17 donors. Partial pooling borrows strength across sparse cohorts without discarding data.
- Interpretability — Biotech and pharma need to explain predictions to regulators and clinicians, not just make them. Every parameter has a clear scientific meaning.
“We wanted to draw big conclusions from a big set of data. This is by far the most successful collaboration I’ve seen — and I’ve had many consultants.”