Predicting activity patterns of regulatory elements in zebrafish using ML

The talk is available online too, see booking url for Teams link

During eukaryotic cell division, the genome folds into a three-dimensional structure that defines how regulatory elements, such as enhancers, interact. Furthermore, studies demonstrate that some folding (or mis-folding) patterns of the genome are closely associated to diverse diseases, such as cancer. Therefore, the identification of genome folding patterns has become a highly interesting topic for scientists to interpret how gene regulation is influenced by chromatin architecture formation. However, chromatin folding presents different patterns depending on the cell type and developmental stage of the organism. Here, we introduce three predicting models based on DeepC, an algorithm that uses transfer learning followed by CNNs, able to predict chromatin interactions in short- and long-distance scales from Hi-C datasets of brain, muscle of adult and whole-embryo zebrafish using transcriptional data alone.