Skip to main content
Fig. 4 | BMC Bioinformatics

Fig. 4

From: Dense neural networks for predicting chromatin conformation

Fig. 4

Gradient analysis of probabilities of contact. a The gradient of the probabilities of contact with respect to the surrounding chromatin states was evaluated at every genomic region. As an example, we show a contact map P(cij) together with its respective chromatin states σk (chromosome 2L 9.15–11.55 Mbp). The gradient was evaluated in four different pairs of sites i and j marked as red circles, and it shows how chromatin states should be altered in order to increase the probability of contact between those sites. The gradient profiles thus suggests the activation or inactivation of regions depending on the location of i and j in the contact map and the sequence of chromatin states. b Genome-wide average of probability gradients ∂P(cij)/∂σk sorted by distance of contact between i and j. An increase of inactive states between pairs of sites typically increases the probability of contact, especially at shorter distances of contact (d<300 kbp). An increase of active states on the sites of contact itself typically increases the probability of contact between sites. The inset shows the average squared gradients, which are indicative of the average magnitude of the gradients, therefore highlighting the chromatin regions with the largest weight on determining probability of contact

Back to article page