Prioris.ai Inc. When predictions matter

Services

We focus on building bespoke, interpretable, predictive models for the life science industry. Full list of services include:

Predictive modelling

Safety pharmacology, toxicology, compound ranking/prioritisation, assay optimisation.

Bayesian statistics

Statistical inference for clinical and pre-clinical experiments, mainly from a Bayesian perspective.

Design of Experiments

Maximising the information gained from experiments, avoiding confounding, and obtaining generalisable results.

Automating analytical workflows

Save time, reduce error, increase reproducibility.

Regulatory writing

Specialising in clinical regulatory writing and submissions.

Why Probabilistic Predictive Modelling?

(aka Bayesian machine learning, probabilistic programming, model-based machine learning)

Deal with uncertainty

A principled way to deal with uncertainty in the data, parameters, and models, thereby avoiding over-confident predictions. Uncertainty is propagated and reported for all predictions.

Include external information

Incorporate scientific knowledge and information from databases, experiments, and experts into the structure of the models.

Interpretable models

Models are constructed using background scientific knownledge and are therefore interpretable (we don't throw matrices at algorithms and hope for the best).

Works with small data

When you don't have the luxury of large data, you need models that work with small sample sizes.

Featured work

Predicting drug-induced liver injury (DILI)

First fully Bayesian DILI prediction model is now published in Chemical Research in Toxicology (preprint available HERE). This model has been in production since end 2017 and was short-listed for an AstraZeneca Science Award. The Supplementary material contains code for R/Stan, Python/PyMC3, and Julia/Turing.

Predicting clinical QT-interval prolongation

We used hERG inhibition assay values and clinical Cmax to predict QTc-prolongation in patients. Published in Toxicological Sciences and discusses the benefits of using Bayesian methods for predicting drug safety. [LINK]

A Bayesian causal model to predict the mechanism of organ weight toxicity

Relative organ weights fail to properly adjust for drug-induced changes in body weight. This paper describes a causal modelling approach that appropriately adjusts for changes in body weight, is intuitive to understand, and provides straightforward probabilistic statements of safety. [LINK]

Team

Brynn Kvinlaug, PhD

Co-founder and CEO

Brynn has over a decade of pharmaceutical and CRO experience and leads the strategic clinical regulatory submissions work. She has a Master's and PhD from Cambridge University and has completed four Ironman triathlons.

Stanley E. Lazic, PhD

Co-founder and CSO

Formerly an Associate Director in Statistics and Machine Learning at AstraZeneca, Stan leads the predictive modelling work at Prioris.ai. He has a Master's and PhD from Cambridge University and is author of the book Experimental Design for Laboratory Biologists.

Gabriel Phelan, MSc

Principal Statistician/Data Scientist

Gabe has an MSc in statistics from Simon Fraser University and an independent school teaching license in British Columbia. He is passionate about interpretable model building and the philosophical foundations of probability; his thesis investigated probabilistic topic models.

Scientific Advisory Board

Lorna Ewart, PhD

Chief Scientific Officer, Emulate Inc.

Lorna has twenty years experience in the pharmaceutical industry, mostly in pre-clinical drug safety. She has an honours degree in Pharmacology from the University of Aberdeen and a PhD from Queen Mary University of London. She is a fellow of the Royal Society of Biology and British Pharmacological Society.

Elizaveta Semenova, PhD

Assistant Professor, Imperial College London

Liza is a Lecturer in Biostatistics, Computational Epidemiology, and Machine Learning at Imperial College London. Previously, she was a postdoctoral researcher at the University of Oxford and, prior to that, worked at AstraZeneca, where she developed and applied Bayesian methods for analyzing health data.

Chad Scherrer, PhD

Founder, Informative Prior, LLC

Chad has over ten years of experience in Bayesian analysis. He served as technical lead for language evaluation in DARPA's PPAML program that focused on probabilistic programming languages (PPLs). He has led development of several prototype systems, and is the lead developer for Soss.jl, a PPL implemented in Julia. Chad has a PhD in Mathematics from Indiana University.

News

2024-01-16: Dealing with multiple hypothesis tests

A publication entitled "Why multiple hypothesis test corrections provide poor control of false positives in the real world" is in press at Psychological Methods.

2022-09-22: Presentation at the EMBL-EBI Industry Programme Workshop

A presentation entitled "Ranking targets with desirability functions and latent variable models" was presented at the EMBL-EBI Industry Programme Workshop on Informatics, Genetics and Omics for Target Selection and Validation in Hinxton, UK.

2021-04-12: Presentation at the British Toxicology Congress

Work from Prioris.ai is presented at the Virtual British Toxicology Congress 2021 by AstraZeneca's Dr. Delyan Ivanov. The talk is entitled "Characterising uncertainty within in vitro models for safety assessment in early drug discovery" and is on YouTube.

2020-04-30: Bayesian neural network for toxicity prediction

AstraZeneca and Prioris.ai have developed a Bayesian neural network (BNN) to predict drug-induced liver injury. Standard neural networks need large datasets for training, but BNNs can be trained on much smaller datasets without overfitting. With BNNs we get the flexibility of neural networks and the ability to quantify prediction uncertainty from a Bayesian model. The preprint is available on bioRxiv. UPDATE: The paper is published in Computational Toxicology and was highlighted on the Julia Computing website.

2019-09-25: Prioris.ai is presenting at the SPS 2019 annual meeting

Avid Afzal from AstraZeneca and Stan Lazic from Prioris.ai are presenting a tutorial at the Safety Pharmacology Society's Annual Meeting (23–26 Sept) in Barcelona, Spain. The title is "Predicting Drug-Induced Cardiotoxicity and Demonstrating the Benefits of a Bayesian Approach" and the tutorial will focus on predicting hERG-mediated QT-interval prolongation using probabilistic predictive models.

2019-05-06: Prioris.ai Inc. Founded

Prioris.ai Inc was founded with the aim of using modern Bayesian methods to solve analytical and prediction problems in drug discovery and biomedical research.

Learn More

We build probabilistic predictive models for clients in the pharmaceutical, biotech, and healthcare industries.