Bioinformatic approaches for detection of fusion genes and <i>trans</i>-splicing products

Chimeric genes and transcripts can be biological markers as well as the reasons for tumor progression and development. Modern algorithms and high-throughput sequencing are the complementary clues to the question of the tumor origin and cancer detection as well as to the fundamental question of chimeric genes origin and their influence on molecular processes of the cell. A wide-range of algorithms for chimeric genes detection was developed, with various differences in computing speed, sensitivity, specificity, and focus on the experimental design. There exist three main types of bioinformatic approaches, which act according to the sequencing read length. Algorithms, which focus on short-read high-throughput sequencing (about 50–300 bр of read length) or long-read sequencing (about 5000–100000 bр of read length) exclusively or algorithms, which combine the results of both short and long-read sequencing. These algorithms are further subdivided into: 1) mapping-first approaches (STAR-Fusion, Arriba), which map reads to the genome or transcriptome directly and search the reads supporting the fused gene or transcript; 2) assembly-first approaches (Fusion-Bloom), which assemble the genome or transcriptome from the overlapping reads, and then compare the results to the reference transcriptome or genome to find transcripts or genes not present in the reference and therefore raising questions; 3) pseudoalignment approaches, which do not make local alignment, but just search for the closest transcript subsequence to the reads seed, following the precomputed index for all reference transcripts and provides the results. This article describes the main classes of available software tools for chimeric gene detection, provides the characteristics of these programs, their advantages and disadvantages. To date the most resource intensive and slowest are still assembly-first algorithms. Mapping-first approaches are quite fast and rather accurate at fusion detection, still the fastest and resource-saving are the pseudoalignment algorithms, but, worth noting, that the quick search is carried out at the expense of chimeras search quality decrease.