Hi-cGAN: Prediction of Hi-C interaction matrices with conditional generative adversarial networks
Background: The three-dimensional organization of the genome is a fundamental aspect of its function and regulation. High-throughput chromosome conformation capture techniques, such as Hi-C, have revolutionized our understanding of spatial genome organization. However, 3C-based methods are resource-intensive and technically demanding. This has driven the development of computational approaches…
High-throughput chromosome conformation capture techniques, like Hi-C, have transformed our understanding of the three-dimensional organization of the genome and its regulation. However, these methods are resource-intensive and technically demanding, leading to the development of computational approaches for predicting Hi-C interaction matrices.
A novel approach called Hi-cGAN has emerged, utilizing conditional generative adversarial networks to provide a computational alternative to extensive wet-lab work. This computational method contributes to a deeper exploration and comprehension of genome architecture.
The Hi-cGAN network employs a convolutional generator paired with a convolutional discriminator, evaluated across various bin sizes ranging from 2 to 25 kb. It generates a whole genome in the form of a cool file, capturing interactions at different resolutions. Notably, the network can emit fixed windows of 1 Mb, 2 Mb, and 990 kb, corresponding to bin sizes of 2 kb, 5-10 kb, and 25 kb, respectively.
When tested on a validation chromosome, Hi-cGAN's agreement with existing methods, such as Epiphany, C.Origami, and Akita, is promising. At a mean correlation of 0.238 across 411 held-out windows, Hi-cGAN's performance aligns closely with Epiphany's results and remains below the more demanding C.Origami and Akita methods.
Boundary and loop calls, which capture larger-scale chromatin organization, demonstrate slightly less agreement with Hi-cGAN-generated maps, placing them at a slightly coarser domain scale. The most informative chromatin factor track varies depending on the resolution: CTCF and cohesin subunits play significant roles at 5-10 kb, while active histone marks become crucial at 25 kb.
Transferring Hi-cGAN's predictions to an unseen cell type incurs a computational cost of approximately 0.12 SCC (Standardized Cell Complexity). The choice between Hi-cGAN and other methods depends on the specific measure being used to evaluate the results.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
