Search for: All records

Understanding genomic sequences through the lens of language modeling has the potential to revolutionize biological research, yet challenges in tokenization, model architecture, and adaptation to diverse genomic contexts remain. In this study, we investigated key innovations in DNA sequence modeling, treating DNA as a language and applying language models to genomic data. We gathered two diverse pretraining datasets: one consisting of 19,551 reference genomes, including over 18,000 prokaryotic genomes (115B nucleotides), and another more balanced dataset with 1,354 genomes, including 1,166 prokaryotic and 188 eukaryotic reference genomes (180B nucleotides). We trained five byte-pair encoding tokenizers and pretrained 52 DNA language models, systematically comparing different architectures, hyperparameters, and classification heads. We introduceseqLens, a family of models based on disentangled attention with relative positional encoding, which outperforms state-of-the-art models in 13 of 19 benchmarking phenotypic predictions. We further explore continual pretraining, domain adaptation, and parameter-efficient fine-tuning methods to assess trade-offs between computational efficiency and accuracy. Our findings demonstrate that relevant pretraining data significantly boosts performance, alternative pooling techniques enhance classification, and larger tokenizers negatively impact generalization. These insights provide a foundation for optimizing DNA language models and improving genome annotations. 

Abstract Proteins are direct products of the genome and metabolites are functional products of interactions between the host and other factors such as environment, disease state, clinical information, etc. Omics data, including proteins and metabolites, are useful in characterizing biological processes underlying COVID-19 along with patient data and clinical information, yet few methods are available to effectively analyze such diverse and unstructured data. Using an integrated approach that combines proteomics and metabolomics data, we investigated the changes in metabolites and proteins in relation to patient characteristics (e.g., age, gender, and health outcome) and clinical information (e.g., metabolic panel and complete blood count test results). We found significant enrichment of biological indicators of lung, liver, and gastrointestinal dysfunction associated with disease severity using publicly available metabolite and protein profiles. Our analyses specifically identified enriched proteins that play a critical role in responses to injury or infection within these anatomical sites, but may contribute to excessive systemic inflammation within the context of COVID-19. Furthermore, we have used this information in conjunction with machine learning algorithms to predict the health status of patients presenting symptoms of COVID-19. This work provides a roadmap for understanding the biochemical pathways and molecular mechanisms that drive disease severity, progression, and treatment of COVID-19.