Hyperdimensional computing: a fast, robust and interpretable paradigm for biological data

TL;DR

This paper addresses the limitations of deep learning in bioinformatics—namely, data hunger and limited interpretability—by advocating hyperdimensional computing (HDC) as a fast, interpretable alternative. It introduces hypervectors and a small set of operations (generating, bundling, binding, permutation) to encode and manipulate complex biological concepts, and surveys encoding strategies for sequences, graphs, and omics data while outlining learning workflows. The authors articulate four major opportunities for bioinformatics: fast, efficient computation; explainability through reversible operations; seamless multimodal data fusion; and symbolic, hierarchical representations that support structured reasoning, including potential in phylogeny and genetic engineering. They argue that HDC can complement deep learning, enabling scalable, explainable analyses across diverse data types, with hardware-aware implementations and hybrid neuro-symbolic models as promising directions. Overall, HDC offers a lightweight, versatile paradigm that can augment current DL approaches for faster, more transparent bioinformatics analyses across omics, biosignals, and health applications.

Abstract

Advances in bioinformatics are primarily due to new algorithms for processing diverse biological data sources. While sophisticated alignment algorithms have been pivotal in analyzing biological sequences, deep learning has substantially transformed bioinformatics, addressing sequence, structure, and functional analyses. However, these methods are incredibly data-hungry, compute-intensive and hard to interpret. Hyperdimensional computing (HDC) has recently emerged as an intriguing alternative. The key idea is that random vectors of high dimensionality can represent concepts such as sequence identity or phylogeny. These vectors can then be combined using simple operators for learning, reasoning or querying by exploiting the peculiar properties of high-dimensional spaces. Our work reviews and explores the potential of HDC for bioinformatics, emphasizing its efficiency, interpretability, and adeptness in handling multimodal and structured data. HDC holds a lot of potential for various omics data searching, biosignal analysis and health applications.

Figures (2)

• Figure 1: a) The hallmarks of hyperdimensional computing (HDC). Hypervectors (HVs) work reliably due to their large dimensionality $N$ (i.e., the Law of Large Numbers states that element-wise properties $S_N$, such as the fraction of positive elements, converge to their expected value for large $N$), and the space is very homogeneous (e.g., most HVs are approximately equidistant). The information about an object is encoded holographically, and the information is robust to random errors. b) Overview of the elementary operations of hyperdimensional computing (HDC): generating, bundling, binding, and shifting. c) Similarity is computed based on element-wise comparisons. d) General HDC workflow, based on Thomas and Rosing Thomas2021theoryofHDC, where red boxes indicate the data space and blue boxes indicate operations in the hyperdimensional space.
• Figure 2: Opportunities for bioinformatics. i) HDC is computationally efficient because it can usually be done using simple bit or arithmetic operations, ii) it is explainable because of its reversibility, iii) it can easily combine different types of data sources, and iv) it can represent complex, structured and hierarchical information.