DNA sequence analysis landscape: a comprehensive review of DNA sequence analysis task types, databases, datasets, word embedding methods, and language models

Muhammad Nabeel Asim, Muhammad Ali Ibrahim, Alam Zaib, Andreas DengelFrontiers in Medicine 2025 · DOI

TL;DR

This paper provides a comprehensive review of the DNA sequence analysis landscape, bridging the gap between computational genomics and AI. It categorizes 44 distinct DNA analysis tasks, surveys relevant databases and datasets, and details the application of word embeddings and language models, offering a valuable resource for researchers in both fields.

What Problem Does This Paper Address?

The rapid advancement of DNA sequencing technologies has generated an overwhelming amount of genomic data. Analyzing this data is crucial for understanding biological processes, diagnosing diseases, and developing targeted therapies. However, a significant gap exists between the biological understanding of DNA analysis tasks and the computational expertise required to develop effective AI solutions. Traditional wet-lab methods are costly and time-consuming, making AI-driven approaches essential for accelerating discovery. This paper aims to bridge this gap by providing a unified overview of DNA sequence analysis tasks, AI methodologies, and the resources needed to develop and evaluate AI predictors.

Methods

The paper's core contribution lies in its comprehensive review and categorization of the DNA sequence analysis landscape. The authors:

  • Categorize 44 DNA sequence analysis tasks: These tasks are grouped into 8 major biological goals (e.g., genome structure and stability, gene expression regulation, disease analysis) and further classified computationally into regression, classification (binary, multi-class, multi-label), and clustering.
  • Survey AI methodologies: They detail the application of 39 word embedding methods (e.g., Word2Vec, FastText, GloVe, Node2Vec) and 67 language models (e.g., BERT, GPT-3, Transformer-XL) in DNA sequence analysis.
  • Analyze predictors: The paper reviews machine learning (e.g., SVM, Random Forest), deep learning (e.g., CNN, LSTM, GNN), and statistical algorithms used in predictive pipelines.
  • Evaluate performance metrics: A detailed overview of evaluation criteria for binary/multi-class classification, multi-label classification, regression, and clustering tasks is provided.
  • Examine open-source resources: The authors compile lists of publicly accessible databases and open-source predictive models, including their performance metrics.

Datasets

The paper identifies and discusses a vast array of datasets crucial for AI-driven DNA sequence analysis:

  • 36 publicly accessible biological databases: These databases provide raw DNA sequences and related data, categorized by volume of sequences (low, medium, high facilitators). Examples include Descartes, COSMIC, and GenBank.
  • 242 benchmark datasets: Out of 127 studies reviewed, 199 are publicly available and 43 are in-house datasets used for various DNA sequence analysis tasks.
  • 140 benchmark datasets: Specifically highlighted for performance comparisons between new and existing AI predictors.
  • Table 2: Provides an overview of 199 public and 43 in-house datasets used across 44 DNA sequence analysis tasks, detailing their use in language models, word embeddings, and other methods.

Key Insights and Ideas

  • The "domain gap" is a major bottleneck: The paper emphasizes the need to bridge the understanding gap between computational scientists and molecular biologists to accelerate AI adoption in genomics.
  • Representation learning is key: The effectiveness of AI in DNA sequence analysis hinges on sophisticated methods for transforming raw sequences into meaningful numerical representations (word embeddings and language models).
  • Standardization is urgently needed: The current landscape suffers from a lack of standardized datasets and evaluation protocols, hindering fair performance comparisons and reproducibility.
  • Untapped potential exists: The paper identifies numerous unexplored word embedding methods, language models, and predictor algorithms that could significantly advance the field.
  • Task categorization is crucial: Classifying tasks both biologically and computationally helps researchers select appropriate AI tools and develop more specialized solutions.

Key Results

  • Comprehensive cataloging: The paper provides a structured overview of 44 distinct DNA sequence analysis tasks, 36 biological databases, 140 benchmark datasets, 39 word embedding methods, and 67 language models.
  • Performance insights: Tables 7-12 summarize the performance of various predictive pipelines across different tasks, highlighting state-of-the-art results and areas for improvement. For instance, in genome structure and stability tasks, BERT and Transformer models with self-classifiers or LSTM classifiers often achieve top performance.
  • Open-source resource identification: The paper compiles lists of publicly available source codes for 25 word embedding-based and 38 language model-based predictive pipelines, facilitating reproducibility and further research.

Limitations and Future Work

  • Limited exploration of advanced methods: The paper notes that many promising word embedding methods (e.g., DANE, ELMO) and language models (e.g., AlphaFold, ESM-2) are yet to be applied to DNA sequence analysis. Similarly, 24 predictor algorithms remain largely unexplored.
  • Need for standardized benchmarks: The authors stress the urgent need for standardized datasets and evaluation protocols to enable fair comparisons and accelerate progress.
  • Potential for multi-task learning: Future research should focus on developing frameworks that can handle multiple DNA sequence analysis tasks simultaneously.
  • Interdisciplinary collaboration: Deeper collaboration between AI researchers and geneticists/bioinformaticians is essential to enhance the biological relevance and impact of AI-driven DNA sequence analysis.
  • Improving underperforming tasks: Specific tasks like disease risk estimation and genome structure analysis show room for improvement, suggesting avenues for future research with advanced AI architectures and ensemble methods.

Summary updated

Abstract

Deoxyribonucleic acid (DNA) serves as fundamental genetic blueprint that governs development, functioning, growth, and reproduction of all living organisms. DNA can be altered through germline and somatic mutations. Germline mutations underlie hereditary conditions, while somatic mutations can be induced by various factors including environmental influences, chemicals, lifestyle choices, and errors in DNA replication and repair mechanisms which can lead to cancer. DNA sequence analysis plays a pivotal role in uncovering the intricate information embedded within an organism's genetic blueprint and understanding the factors that can modify it. This analysis helps in early detection of genetic diseases and the design of targeted therapies. Traditional wet-lab experimental DNA sequence analysis through traditional wet-lab experimental methods is costly, time-consuming, and prone to errors. To accelerate large-scale DNA sequence analysis, researchers are developing AI applications that complement wet-lab experimental methods. These AI approaches can help generate hypotheses, prioritize experiments, and interpret results by identifying patterns in large genomic datasets. Effective integration of AI methods with experimental validation requires scientists to understand both fields. Considering the need of a comprehensive literature that bridges the gap between both fields, contributions of this paper are manifold: It presents diverse range of DNA sequence analysis tasks and AI methodologies. It equips AI researchers with essential biological knowledge of 44 distinct DNA sequence analysis tasks and aligns these tasks with 3 distinct AI-paradigms, namely, classification, regression, and clustering. It streamlines the integration of AI into DNA sequence analysis tasks by consolidating information of 36 diverse biological databases that can be used to develop benchmark datasets for 44 different DNA sequence analysis tasks. To ensure performance comparisons between new and existing AI predictors, it provides insights into 140 benchmark datasets related to 44 distinct DNA sequence analysis tasks. It presents word embeddings and language models applications across 44 distinct DNA sequence analysis tasks. It streamlines the development of new predictors by providing a comprehensive survey of 39 word embeddings and 67 language models based predictive pipeline performance values as well as top performing traditional sequence encoding-based predictors and their performances across 44 DNA sequence analysis tasks.
BibTeX
@inproceedings{asim2025dna,
  author = {Muhammad Nabeel Asim and Muhammad Ali Ibrahim and Alam Zaib and Andreas Dengel},
  title = {{DNA sequence analysis landscape: a comprehensive review of DNA sequence analysis task types, databases, datasets, word embedding methods, and language models}},
  booktitle = {Frontiers in Medicine},
  year = {2025},
  doi = {10.3389/fmed.2025.1503229},
}
Figure