L2M – Developing Machine Learning Models to Accelerate our In-Vivo Creatinine and Potassium Biosensor Development

Many people with heart failure are not prescribed the full set of guideline-directed medical therapies (GDMT), which can shorten their lifespan and reduce quality of life. One major reason is the difficulty and inconvenience of regularly checking potassium and creatinine levels, which are needed to safely adjust these medications. To address this, we are developing the first on-demand system to measure potassium and creatinine, making it possible to optimize GDMT in weeks instead of years.

A key challenge we encountered is interference in electrochemical sensors: when other compounds in biological fluids create signals that interfere with accurate detection of the target molecule. Traditionally, solving this requires slow and labor-intensive trial-and-error testing of electrode materials in the lab. This delays development and limits how quickly new sensors can be brought to use.

Our solution is to use existing data and electrochemical knowledge to train machine learning models that can predict potential interferents and recommend better sensor designs before lab testing. This approach transforms biosensor development into a faster, data-driven process and could significantly accelerate the creation of reliable, clinically useful sensors.

Faculty Supervisor:

Istvan Mucsi

Student:

Partner:

DMZ Ventures Inc

Discipline:

Life Sciences

Sector:

Professional, scientific and technical services

University:

University of Toronto

Program: