Modeling in vitro selection of binding proteins by Angelo Charry

Currently, directly measuring the physical properties of proteins, such as ligand-affinity and thermal stability, presents a challenge, particularly when seeking to understand the impact of thousands of mutations on these properties. Meanwhile, advancements in new generation sequencing have made high-throughput screening of large mutant libraries more accessible. By utilizing physics-informed machine learning to model these experiments, it becomes possible to determine the effects of mutations on the protein's physical properties, thereby paving the way for sequence engineering.

*************************

Leveraging long-range temporal information for efficient segmentation and tracking by Maxime Deforet

Detecting and tracking objects in videos poses significant challenges. Traditional methods involve segmenting each frame, extracting objects' centers of mass, and linking these across frames, with segmentation often being the most challenging stage. Humans, however, can easily recognize objects while watching videos, suggesting that leveraging temporal context could enhance accuracy. I will present DistNet2D, a new deep neural network architecture for 2D cell segmentation and tracking that leverages long-term temporal information. I will also discuss strategies to generate a training dataset, which is often the bottleneck in supervised learning, including PixMClass, a method for segmentation with sparse annotation.