Branco Weiss Fellow Since
2024
Research Category
Machine Learning, Microfluidics, Biophysics
Research Location
Department of Genetics, Stanford University, Palo Alto, USA
Background
Although all biological functions depend on protein interactions, we still cannot systematically predict the binding affinity of two protein sequences. Protein-protein interactions (PPIs), especially those related to cell signaling, are often weak and transient, making them difficult to measure. While advances in machine learning have revolutionized protein folding, applying these models to predict protein binding has produced inconsistent results since the training data and models are not tailored for PPIs. Therefore, developing machine learning algorithms that accurately predict PPIs will require new model architectures and, crucially, a new kind of data.
Details of Research
Dr. Karl Krauth’s background as a machine learning researcher embedded in a microfluidics laboratory uniquely positions him to tackle the protein interaction problem. In his project, Dr. Krauth will develop a platform that integrates valved microfluidic technology, cDNA display, and next-generation sequencing to quantify the binding strength of over ten million distinct PPIs in a single experiment. This technology will enable him to collect the largest and most detailed PPI dataset to date, reporting thermodynamic and kinetic quantities across protein families spanning the proteome. He will use this dataset to train a deep learning model capable of predicting PPIs directly from sequences. By incorporating an active learning algorithm, he will iteratively refine the model through subsequent rounds of machine-learning-guided data collection.
