CacPred: a cascaded convolutional neural network for TF-DNA binding prediction

• DOI: 10.1186/s12864-025-11399-y
• Publication Date: 2025-03-18T09:39:34Z
• Authors: Shuangquan Zhang; Anjun Ma; Xuping Xie; Zhichao Lian; Yan Wang
• Keywords: Chromatin immunoprecipitation

Abstract Background Transcription factors (TFs) regulate the genes’ expression by binding to DNA sequences. Aligned TFBSs of same TF are seen as cis-regulatory motifs, and substantial computational efforts have been invested find motifs. In recent years, convolutional neural networks (CNNs) succeeded in TF-DNA prediction, but existing DL methods’ accuracy needs be improved convolution function prediction should further explored. Results We develop a cascaded network model named CacPred predict on 790 Chromatin immunoprecipitation-sequencing (ChIP-seq) datasets seven ChIP-nexus (chromatin immunoprecipitation experiments with nucleotide resolution through exonuclease, unique barcode, single ligation) datasets. compare six models across nine standard evaluation metrics. Our results indicate that outperforms all comparison for average (ACC), matthews correlation coefficient (MCC), area eight metrics radar (AEMR) 3.3%, 9.2%, 6.4% ChIP-seq Meanwhile, improves ACC, MCC, AEMR 5.5%, 16.8%, 12.9% To explain proposed method, motifs used show features learned. light results, can some significant from input Conclusions This paper indicates performs better than data. Seven also analyzed, they coincide our method best only is equipped algorithm, demonstrating pooling processing leads losing sequence information. Some found, showing learn this study, we demonstrate an effective feasible predicting binding. freely available at https://github.com/zhangsq06/CacPred .