Single-Cell RNA sequencing technology provides an opportunity to discover gene regulatory networks(GRN) that control cell differentiation and drive cell type transformation. However, it is faced with the challenge of high loss and high noise of sequencing data and contains many pseudo-connections. To solve these problems, we propose a framework called Fusion prior gene network for Gene Regulatory Network inference Accuracy Enhancement(FGRNAE) to infer a high reliable gene regulatory network. Specifically, based on the Single-Cell RNA-sequencing Network Propagation and network Fusion(scNPF) preprocessing framework, we employ the Random Walk with Restart on the prior gene network to interpolate the missing data. Furthermore, we infer the network using the Random Forest algorithm with the results achieved above. In addition, we apply data from the Co-Function Network to build a meta-gene network and select the regulatory connection with the Markov Random Field. Extensive experiments based on datasets from BEELINE validate the effectiveness of our framework for improving the accuracy of inference.

Imbalanced data classification is the fundamental problem of data mining. Relevant researchers have proposed many solutions to solve the problem, such as sampling and ensemble learning methods. However, random under-sampling is easy to lose representative samples, and ensemble learning does not use the correlation information between pieces in the data set. Therefore, we proposed a Hybrid Adaptive sampling with Bagging Classifier(HABC). Specifically, we calculated the adaptive sampling rate according to the characteristics of the data set. We then performed density-based under-sampling and over-sampling on the original data set according to the sampling rate. Further, the sampled data subset was sent to the Bagging classifier, and the classifier was employed to predict the unknown data set. In addition, the multi-objective particle swarm optimization algorithm was combined to optimize the prediction result. Extensive experiments based on UCI, KEEL, and three bioinformatics datasets show that our proposed method is better than state-of-the-art algorithms.

To comprehend DNA-protein binding specificity in diverse cell types is essential to reveal regulatory mechanisms in biological processes. Recently, deep learning has been successfully applied to predict DNA-protein binding sites from large-scale chromatin-profiling data. However, the precise identification of putative binding sites in specific cell types with low labeled samples remains challenging. To this end, we present a novel Transferable Transformer-based method, dubbed as BindTransNet, for cross-cell types DNA-protein binding prediction. Transfer learning and Transformer Encoder are simultaneously adopted in our presented approach to capture some shared long-range dependencies between various motifs available in cross-cell types. This unique design helps our method recognize putative binding sites without massive labeled samples by leveraging the above-mentioned standard features. This work is the first to apply a Transformer for DNA-protein binding sites prediction. The presented method is measured on TFs COREST and SRF in four cell types with eight cell-type TF pairs. For both 4-class prediction and binary-level prediction, BindTransNet can significantly outperform several state-of-the-art methods. Moreover, BindTransNet achieves considerable margin performance improvements by leveraging transfer learning. This is a presuasive indication that BindTransNet can indeed capture shared features available in other cell types.