Integrating Natural Language Processing techniques to enhance genomic sequence analysis and characterization

The vast amount of data in biological science has increased in the last year, leading to advances in various areas. Natural language processing (NLP) has demonstrated its capability to analyze large volumes of text and automatically extract relevant information. The use of NLP can be particularly valuable in facilitating the identification of cells, bacteria or viruses from the analysis of genetic code sequences. However, the genomic data incorporation to this analysis is still in early stages. Data from different sources and sequencing platforms need to be integrated and normalized, which is resource-intensive. The complexity of these tasks increases exponentially with the size of the data, making it impractical to use conventional computational approaches. The objective of this study is to explore how the integration of NLP algorithms can help to improve the identification of organisms and their characteristics. To this end, we utilize DNABERT, a modified version of BERT, a transformer-based contextualized language representation model. We apply this model to a dataset of more than 454,592 genomic sequences from patients infected with various respiratory conditions, including Influenza, COVID-19, and Aspergillosis-associated coinfections. The data was acquired from theNational Center for Biotechnology Information (NCBI). Aspergillosis, a fungal infection caused by Aspergillus species, is particularly concerning in patients with pre-existing respiratory diseases or compromised immune systems, such as those suffering from severe Influenza or COVID-19. By leveraging DNABERT, a specialized deep learning model that tokenizes DNA sequences into smaller k-mer segments, the study was able to identify unique sequence patterns and motifs across these conditions. The results demonstrated the model’s capacity to differentiate between various respiratory conditions by capturing thecontextual dependencies within the genomic data. For example, specific k-mers were identified as being highly associated with COVID-19 or Influenza, which provides a deeper understanding of the genetic factors contributing to disease manifestation and progression. Additionally, the clustering of sequences based on their k-mer profiles revealed distinct groups for each condition, while also showing some degree of overlap between COVID-19 and COVID-19-associated pulmonary aspergillosis. This suggests potential biologicalinteractions or shared characteristics between these infections, which may have implications for diagnosis and treatment strategies. uture work will focus on expanding the model to incorporate additional metadata, such as patient demographics or clinical outcomes, and exploring advanced techniques like transfer learning to enhance predictive performance and uncover subtle patterns within the data. Acknowledgement: Departamento de Informática y Computación, UTEM; Escuela de Informática, UTEM; Laboratorio de Investigación Aplicada, Departamento de Informática y Computación, UTEM. This work was supported in part by Project supported by the “Competition for Research Regular Projects”, year 2023, codeLPR23-09 and “Competition for Research Assistant Funding UTEM”, year 2023, code AI23-06, Universidad Tecnológica Metropolitana (AM-B)