Date: 9 Dec 2025
Time: 4pm GMT | 11am EST | 8am PT | 6pm CEST
Accurate QSAR models lead to more efficient and cost-effective molecular discovery. Better predictions enable you to prioritise the optimal compounds for your projects, reducing synthesis and testing requirements. However, the challenge lies in diversity: how do we build models that consistently perform across such a wide range of molecular property prediction tasks?
In this webinar, we’ll explore our latest research, published in JCIM, introducing our MetaModel framework that addresses exactly this.
Our approach combines two key steps. First, we use graph neural network featurisation, combining task-specific learned molecular descriptors with general-purpose descriptors. This ‘best of both’ strategy ensures that we have the best data on which to train our models. Second, we integrate predictions from a diverse set of machine learning algorithms into a single consensus model, applying the most suitable techniques for every problem.
Together, these steps deliver more accurate and reliable QSAR models that demonstrate consistent performance across diverse problem types. MetaModel outperforms leading approaches like ChemProp, delivering substantial improvements where traditional neural networks struggle, whilst matching or exceeding performance where they already excel.
