Pairwise Heuristic Sequence Alignment Algorithm Based on Deep Reinforcement Learning.

Goal: Various methods have been developed to analyze the association between organisms and their genomic sequences. Among them, sequence alignment is the most frequently used method for comparative analysis of biological genomes. We intend to propose a novel pairwise sequence alignment method using deep reinforcement learning to break out the old pairwise alignment algorithms. Methods: We defined the environment and agent to enable reinforcement learning in the sequence alignment system. This novel method, named DQNalign, can immediately determine the next direction by observing the subsequences within the moving window. Results: DQNalign shows superiority in the dissimilar sequence pairs that have low identity values. And theoretically, we confirm that DQNalign has a low dimension for the sequence length in view of the complexity. Conclusions: This research shows the application method of deep reinforcement learning to the sequence alignment system and how deep reinforcement learning can improve the conventional sequence alignment method.

Genome classification improvements based on k-mer intervals in sequences.

Given the vast amount of genomic data, alignment-free sequence comparison methods are required due to their low computational complexity. k-mer based methods can improve comparison accuracy by extracting an effective feature of the genome sequences. The aim of this paper is to extract k-mer intervals of a sequence as a feature of a genome for high comparison accuracy. In the proposed method, we calculated the distance between genome sequences by comparing the distribution of k-mer intervals. Then, we identified the classification results using phylogenetic trees. We used viral, mitochondrial (MT), microbial and mammalian genome sequences to perform classification for various genome sets. We confirmed that the proposed method provides a better classification result than other k-mer based methods. Furthermore, the proposed method could efficiently be applied to long sequences such as human and mouse genomes.

Repetitive element signature-based visualization, distance computation, and classification of 1766 microbial genomes.

The genomes of living organisms are populated with pleomorphic repetitive elements (REs) of varying densities. Our hypothesis that genomic RE landscapes are species/strain/individual-specific was implemented into the Genome Signature Imaging system to visualize and compute the RE-based signatures of any genome. Following the occurrence profiling of 5-nucleotide REs/words, the information from top-50 frequency words was transformed into a genome-specific signature and visualized as Genome Signature Images (GSIs), using a CMYK scheme. An algorithm for computing distances among GSIs was formulated using the GSIs' variables (word identity, frequency, and frequency order). The utility of the GSI-distance computation system was demonstrated with control genomes. GSI-based computation of genome-relatedness among 1766 microbes (117 archaea and 1649 bacteria) identified their clustering patterns; although the majority paralleled the established classification, some did not. The Genome Signature Imaging system, with its visualization and distance computation functions, enables genome-scale evolutionary studies involving numerous genomes with varying sizes.

Genome-based microorganism classification using coalition formulation game.

Genome-based microorganism classification is the one of interesting issues in microorganism taxonomy. However, the advance in sequencing technology requires a low-complex algorithm to process a great amount of bio sequence data. In this paper, we suggest a coalition formation game for microorganism classification, which can be implemented in distributed manner. We extract word frequency feature from microorganism sequences and formulate the coalition game model that considers the distance among word frequency features. Then, we propose a coalition formation algorithm for clustering microorganisms with feature similarity. The performance of proposed algorithm is compared with that of conventional schemes by means of an experiment. According to the result, we showed that the correctness of proposed distributed algorithm is similar to that of conventional centralized schemes.

The glycosylation and in vivo stability of human granulocyte-macrophage colony-stimulating factor produced in rice cells.

Recombinant human granulocyte-macrophage colony-stimulating factor (rhGM-CSF) has been produced in several different cell types, and has different properties depending on the production process used. There has been no report of glycosylation ratio or of the role of glycans for plant cells-derived rhGM-CSF. We investigated the terminal sialylation of rhGM-CSF produced in rice cells (rrhGM-CSF) using lectins. Glycosylation ratios of rrhGM-CSF and yeast-derived rhGM-CSF (yrhGM-CSF) were evaluated by chemical deglycosylation. After intravenous administration to rats, the clearance rates of rrhGM-CSF and yrhGM-CSF as well as deglycosylated forms of them were evaluated. In vitro bioactivities of rrhGM-CSF and yrhGM-CSF were measured in dose- and time-dependent manners. Although rrhGM-CSF does not have terminal sialic acids, the glycosylation ratio was two-fold higher than that of yrhGM-CSF, and rrhGM-CSF sustained longer than yrhGM-CSF in the blood circulation. Moreover, yrhGM-CSF and rrhGM-CSF have almost same bioactivity via dose- and time-dependent manners. Deglycosylated rrhGM-CSF was cleared faster than native forms, while deglycosylated yrhGM-CSF and yrhGM-CSF were similarly cleared in blood circulation. These results suggest that the glycans on rrhGM-CSF are superior to the glycans on yrhGM-CSF in terms of in vivo stability.