The interpretation of our genetic code is largely managed by a set of proteins that specialize in finding where the genes in our DNA are located, and bind to them. Then, in a complex network of interactions, these specialized proteins coordinate the expression of our genes into functional elements in our body. What we know for sure is that in order for biology to be this consistent, these proteins must have very specific binding preferences for particular sequences of DNA. However, recent studies have shown that the sequence itself cannot single-handedly explain this binding specificity. Instead, the probability of a protein binding to its sequence is affected by the influence of surrounding DNA regions and, in some cases, whether another protein is present to help or obstruct the binding act. By combining state-of-the-art techniques in machine learning and high-throughput measurements, we're able to model and predict much more complicated interaction patterns than ever before. This enables us to peek inside this complex interaction network and gain quantitative understanding of these phenomena. This not only has the potential to contribute to the field of genomics, but also helps guide research in fields such as medicine toward better treatment of genetic disorders and diseases driven by mutations, such as cancer.

Warnerfjord, M. (2025). Predicting Protein-DNA Binding Affinity Using AlphaFold Embeddings & Graph Attention Networks.