Biological Sequence Clustering: A Survey

TL;DR

This survey addresses the escalating scale and diversity of biological sequence data by organizing clustering methods into a three-layer framework: similarity modeling, clustering mechanisms, and algorithm design objectives. It unifies alignment-based and alignment-free approaches, detailing sequence encoding, feature generation, and similarity measurement, and then classifies clustering methods into seven paradigms with emphasis on trade-offs between scalability, interpretability, and robustness. The authors provide a comprehensive synthesis of classic and emerging techniques—ranging from greedy and hierarchical to graph-based, model-based, and deep-learning approaches—and discuss practical design considerations, including post-processing refinements and domain-specific interpretability. The work highlights key challenges, such as scalability to ultra-large datasets, reliability under noise, and maintaining biological relevance at low sequence similarity, and offers guidance for future method development and application in large-scale, heterogeneous sequence analyses.

Abstract

The rapid development of high-throughput sequencing technologies has led to an explosive increase in biological sequence data, making sequence clustering a fundamental task in large-scale bioinformatics analyses. Unlike traditional clustering problems, biological sequence clustering faces unique challenges due to the lack of direct similarity measures, strict biological constraints, and demanding requirements for both scalability and accuracy. Over the past decades, a wide variety of methods have been developed, differing in how they model sequence similarity, construct clusters, and prioritize optimization objectives. In this review, we provide a comprehensive methodological overview of biological sequence clustering algorithms. We begin by summarizing the main strategies for modeling sequence similarity, which can be divided into three stages: sequence encoding, feature generation, and similarity measurement. Next, we discuss the major clustering paradigms, including greedy incremental, hierarchical, graph-based, model-based, partitional, and deep learning approaches, highlighting their methodological characteristics and practical trade-offs. We then discuss clustering objectives from three key perspectives: scalability and resource efficiency, biological interpretability, and robustness and clustering quality. Organizing existing methods along these dimensions allows us to explore the trade-offs in biological sequence clustering and clarify the contexts in which different approaches are most appropriate. Finally, we identify current limitations and challenges, providing guidance for researchers and directions for future method development.

Figures (3)

• Figure 1: A methodological overview of biological sequence clustering.
• Figure 2: Similarity modeling framework for biological sequences. The analytical pipeline is organized into three steps: sequence encoding, feature generation, and similarity measurement. Given two input DNA sequences (S1: ATGCG and S2: ATGAG), the framework branches into alignment-based and alignment-free paradigms. Alignment-based methods operate on raw character encodings and compute similarity directly through sequence identity or alignment scores. In contrast, alignment-free approaches either apply substring-based encoding to derive word-based features, such as 3-mer frequency vectors (S1=(1,1,1,0,0), S2=(1,0,0,1,1)) or sketches, or extract information-based features, for example, Shannon entropy and sequence complexity. Similarity is then quantified using a range of metrics, such as Euclidean distance, Jaccard similarity, complexity, and Kullback–Leibler divergence. The resulting similarity information is ultimately represented as a similarity matrix or a weighted graph, which serves as the input for subsequent clustering analyses.
• Figure 3: Comparison between greedy incremental-based and graph-based clustering mechanisms. (Left) Greedy incremental-based strategies (e.g., CD-HIT, UCLUST) partition sequences into clusters using central seeds. This local decision-making often overlooks connections between clusters, which can fragment continuous evolutionary lineages. (Right) Graph-based methods define clusters based on overall connectivity. By capturing the connected relationship (e.g., A–B–C–D), these approaches can recover a continuous sequence relationship that connects divergent members through multiple intermediate sequences, effectively resolving evolutionary structures.