VP of Engineering
Researcher / Engineer
Cancer research with AI methods
Developing a new drug on average costs well over $2 billion and takes more than 10 years. Turbine.AI was founded to bring cutting edge technology to this field, leveraging AI models built on top of the vast amount of publicly available data points. Through this, we optimize both the costs and the timeline for creating novel treatments.
Our talk will firstly walk you through the main challenges of this field, like data quality and the need for interpreting our model behavior – by processing terabytes of data per simulation. We will also cover how we are building the organization to cope well with the rapid introduction of new technologies in a highly multidisciplined team setup.
The second part will focus on how we do research at the company. While having a stable platform is crucial for daily operations, a dedicated group of researchers work on the next generation of our cell model. Fusing ML with cancer cell biology is a constant challenge – there are no low hanging fruits, easy to re-use algorithms in this area. We’ll talk about how to experiment and build working prototypes in such a high risk – high reward environment. We’ll also discuss how to apply the latest advancements of ML in the world of noisy (frequently unlabelled) biological data full of heavily out-of-distribution (OOD) test sets. And how can we utilize graph machine learning (GNNs) for modeling protein-protein interaction (PPI) networks.
Leading AI research at Turbine, Gabor is responsible for delivering the next generation of the company’s core technology, the Simulated Cell model. This involves applying deep learning techniques in the field of cancer cell biology. A domain-heavy environment like that, requires novel model architectures to be developed, that can scale-out, yet keep interpretability. His latest research is focusing on graph ML / geometric deep learning on large datasets.