Unlocking the Predictive Power of Machine Learning for Biotherapeutic Development

Proteins are complex molecules! Even the well-studied ones, like monoclonal antibodies (mAbs), often behave unpredictably. To accelerate the development of biotherapeutics, Artificial Intelligence (AI) and Machine Learning (ML) promise better predictions of protein behavior. However, to unlock the predictive power of these new technologies we still need the domain knowledge of expert scientists. Their expertise comes to augment ML approaches in two ways. Firstly, they can guide the development of ML by asking the fundamental questions underpinning molecular behavior. Secondly, they can provide the interpretation of results and this opens the ‘black box’ and leads to explainable AI.

At Malvern Panalytical we set up a Data Factory so that we can compile datasets that would help us develop ML models. We focused on three techniques that are commonly used for the stability and developability assessment of mAbs. These are Differential Scanning Calorimetry (DSC), Dynamic Light Scattering (DLS), and Size Exclusion Chromatography (SEC). Once we had the data in our hands, we employed advanced analytics to extract unique insights about the mAbs we characterized. We uncovered ways of combining data in analytical workflows that help to better discriminate and explain the behavior of different mAbs, and we investigated ML models that have the capability to predict their biophysical attributes.

How did we do that?

To find out watch the short presentations below that our team recently delivered at the PEGS Europe 2021 conference (Barcelona, 2-4 November).