Kamil Luczkiewicz

Knotted topologies in biopolymers and proteins are a fascinating yet challenging phenomenon with growing relevance in structural biology and chemistry. These complex molecular shapes are increasingly recognized for their potential roles in biological function, but traditional mathematical tools often fall short in detecting and classifying them due to knot theory limitations. Recent advances in machine learning, particularly in sequence-based models like LSTMs and attention mechanisms, offer new possibilities for analyzing non-trivial topologies. Using only 3D structural data, ML models can learn to recognize non-trivial topologies, like knots, theta-curves or lassos, showing the possibilities of making the process easier and faster than classical algorithmic methods.