A deep learning framework combined with word embedding to identify DNA replication origins

Abstract

The DNA replication influences the inheritance of genetic information in the DNA life cycle. As the distribution of replication origins (ORIs) is the major determinant to precisely regulate the replication process, the correct identification of ORIs is significant in giving an insightful understanding of DNA replication mechanisms and the regulatory mechanisms of genetic expressions. For eukaryotes in particular, multiple ORIs exist in each of their gene sequences to complete the replication in a reasonable period of time. To simplify the identification process of eukaryote’s ORIs, most of existing methods are developed by traditional machine learning algorithms, and target to the gene sequences with a fixed length. Consequently, the identification results are not satisfying, i.e. there is still great room for improvement. To break through the limitations in previous studies, this paper develops sequence segmentation methods, and employs the word embedding technique, ‘Word2vec’, to convert gene sequences into word vectors, thereby grasping the inner correlations of gene sequences with different lengths. Then, a deep learning framework to perform the ORI identification task is constructed by a convolutional neural network with an embedding layer. On the basis of the analysis of similarity reduction dimensionality diagram, Word2vec can effectively transform the inner relationship among words into numerical feature. For four species in this study, the best models are obtained with the overall accuracy of 0.975, 0.765, 0.885, 0.967, the Matthew’s correlation coefficient of 0.940, 0.530, 0.771, 0.934, and the AUC of 0.975, 0.800, 0.888, 0.981, which indicate that the proposed predictor has a stable ability and provide a high confidence coefficient to classify both of ORIs and non-ORIs. Compared with state-of-the-art methods, the proposed predictor can achieve ORI identification with significant improvement. It is therefore reasonable to anticipate that the proposed method will make a useful high throughput tool for genome analysis.

Genomes contain relics of a triplet code connecting the origins of primordial RNA synthesis to the origins of genetically coded protein synthesis

“…In this study, we considered triplets as words given their role in protein coding genes as codons as well as their possible roles in primordial RNA synthesis as discussed in the introduction. In addition, the use of triplets as words during language-based modeling of genome sequences by previous studies has demonstrated that they capture important biological information that can be used for different prediction tasks such as the inference of DNA replication origins [93], identification of enhancers [94] or transcription factor binding sites [95]. Therefore, to apply word2vec to the metagenomes, each species level metagenome sequence was split into non-overlapping triplets, starting from the 5'-end.…”

“…• First each genome was split into non-overlapping triplets "TAT ACG GGG AAA ...". In previous biological sequence analysis tasks [93,94], overlapping K-mers were used during word embedding. However, in this study the use of overlapping K-mers would introduce a statistical artefact in examining the relationships between adjacent triplets since overlapping triplets obtained by a sliding through the sequence 1-nucleotide a time would be predictably different by 1 nucleotide.…”

Genomic Surveillance of COVID-19 Variants with Language Models and Machine Learning

“…Typically, in language models, word embeddings represent the latent space (dimensions) of a corpus of text (Mikolov et al, 2013) and can capture highly nonlinear and contextual relationships. Codons (tri-nucleotides, 3-mers) translations represent a natural basis for word representations and have been utilized in the past for learning embedding models for modeling various outcomes such as mutation susceptibility (Yilmaz, 2020) and gene sequence correlations (Wu et al, 2021). Recently, Hie et.…”

“…However, there is a paucity of studies that explore the use of unsupervised embeddings for machine learning based prediction of surges in infections. In these models, codons (tri-nucleotides, 3-mers) translations represent a natural basis for word representations and have been utilized in the past for learning embedding models for modelling various outcomes such as mutation susceptibility and gene sequence correlations (Yilmaz, 2020) (Wu et al, 2021). Recently, Hie et.…”

“…Codons with their tri-nucleotide translations represent a biological basis for word representations. They have been utilized to learn embeddings for modeling various outcomes such as mutation susceptibility [6] and gene sequence correlations [7]. Our empirical experiments with the learned embeddings uncovered explainable genomic signals and predicted new caseloads across nine countries.…”